StarTopic Future Nintendo Hardware & Technology Speculation & Discussion |ST| (Read the staff posts before commenting!)

[Donnie]

OFFICIAL STAFF COMMUNICATION

While spirited discussion is absolutely encouraged and enjoyed, there have been a few complaints about the tone of this discussion. The staff have a few key ideas we would like to vocalize to help everyone understand each other correctly and hear each other's points in the way they were intended.

- When expressing a personal view, please try your best to use language that does not present it as objective fact, or that mocks others' thoughts.

- If you are invested enough to engage in the discussion in the first place, please try to respect others' time. Even if you intend only to leave your opinion and move on, please try and back it up and with some thought. If you only have a single line sentence to share, please consider whether it will be a meaningful contribution to the discussion.

- Please try to refrain from attacking reputable news sources. We're all mature enough to engage in discussion about rumors with the understanding that inaccuracies may reveal themselves over time; changes in plans occur, and there are many variables we will never be privy to. Please let staff know if there is a legitimate problem considering a news source, such as a banned or unknown sources.

Most importantly, please try to understand and listen to each other in good faith and try your best to resolve disagreements within the context of the debate. Staff are always here to support you in instances that go further than that. If you find yourself reacting emotionally to another's comment, either bring it to the staff via the report button, or step away from the thread for a little while.

Thank you. And please keep up the great discussion!

 

[Dakhil]

By the way, the Orin X made its first public appearance through Zhiji Motors (a.k.a. IM Motors): "[BGCOLOR=rgb(41, 41, 41)]NVIDIA Orin X chip adopts brand-new NVIDIA GPU and 12-core ARM CPU, made of 7nm process, single-chip computing capacity is up to 254 TOPS per second.[/BGCOLOR] It is at the top level among the current mass-produced automotive-grade AI chips. It is reported that the single-chip computing power of this chip is about 10 times that of Mobileye's latest EyeQ5 and 3.5 times that of Tesla's HW3.0."

I find it interesting that a 7 nm* process was mentioned. And I also find it interesting that there's mention of 'KR' in the labelling for Orin X, which definitely confirms that one of Samsung's sub-14 nm* (11 nm* and below) process nodes are used to fabricate Orin X, since 'KR' probably stands for (South) Korea.

~

Anyway, I've made a poll asking "When do you think is the earliest time Nintendo will launch the DLSS model*?" I've seen so many different predictions from many people here in terms of when the DLSS model* will launch, which made me curious. (And when I mean launch, I mean release.)

* foundry marketing nomenclature

 

[Dakhil]

Speaking about price, I've made a second poll for everyone here to participate in. The poll's about which price people here think Nintendo will likely set as the MSRP for the DLSS model*. Click here to participate in the second poll.

~

And speaking about polls, here are the results for the first poll, as of 26 October 2021.

If you haven't participated in the first poll, you can still do so here.

 

[Look over there]

Also, really interesting discussion regarding the 8nm, 7nm, and on down things! Because I'm dumb, if you were to take an original Switch, which I'm assuming uses 8nm, kept everything the same, but brought it down to 7nm, what kind of performance enhancement comes from that?

Hmm, might as well do a recap, as I'm sure that you're not the only reader wondering 'what's all this nm business in relation to the Switch'
First off, 'x nm' used to literally refer to the length of one aspect of a transistor, but that stopped being the case for a while now (was the decoupling back in the 90's or 2000's?). The nomenclature continues on as a matter of 'eh, might as well'. They're just labels/branding now.
Next, they can generally be organized in terms of generations. I'll refer to the ITRS roadmap from 2017 for generation names.
Generally speaking, each generation should offer significant improvements, while nodes that are just refinements/variants within the same generation should offer minor to moderate improvements.

22nm:
TSMC 20nm node is part of this generation. The original Nintendo Switch (v1, Erista) was manufactured on this node. For consistency with the rest of this post, I'll state that this is 0 generations removed from the first Switch.

14nm:
TSMC 16FF/12FF are part of this generation. Switch-wise, v2/Mariko/Redbox/Lite/OLED. 1 generation removed from the start.
(PS4 Slim/Pro were also made on this node, I think?)
I should point out that the TSMC 20->16FF/12FF jump was huge for power savings, both on paper (probably fair to ballpark it as 2 later generations worth of gains) and in practice, as demonstrated with the v2. There was a major shift in transistor design (the FF stands for FinFET), thus the even better than usual improvement.

10nm:
2 generations removed from the start.
There was a very short lived TSMC 10FF node. On paper, gains weren't shabby actually, but they phased this one out once they got the next generation going.
Samsung's entry here starts with 10 LPE and gets refined over and over. 10LPE->10LPP->8LPP->8LPU. Yes, Samsung's 8nm nodes are just 10nm refinements. Nvidia's Orin is on... one of those Samsung 8 nm nodes. As far as we are aware, Drake's design is an offshoot of Orin, which is why our speculation starts here.

7nm:
3 generations removed from the start.
TSMC's entry here starts with N7, then there are a few variants. N6 is a N7 refinement/variant. On paper, the gains in power efficiency that TSMC advertises for the jump from 16FF/12FF to N7 are... about the same as 20->16FF/12FF. That's more to highlight that wow, 20->16FF/12FF was good. The PS5/Xbox Series consoles are made on this node family. We are aware that Nvidia has/had some capacity on N7 because of the A100 gpu.
Samsung's entry here starts with 7LPP and also gets refined over and over. 7LPP->(6LPP)->5LPE->5LPP->(4LPE). 6LPP is in parentheses because eventually it disappears from presentations. 4LPE is in parentheses because eventually Samsung reclassifies it to being a member of the next generation
Consensus is that TSMC's the better of the two for this generation.

5nm:
4 generations removed from the start. As of today, this is the latest generation with products out on the market.
For TSMC, the N5 and N4 nodes are part of this generation. N5 started production in 2020, N5P in 2021, N4 this year, and N4P's either later this year or next year. The N5 family is one of the suspects since purportedly, Nvidia spent multiple billions of USD to secure capacity on this family for their next generation of products.
For Samsung, 4LPE and 4LPP are officially their entries in this generation.
Again, consensus is that TSMC's the winner of this generation.

3nm:
5 generations removed from the start.
For TSMC, N3 is expected to start volume production by the end of this year, with products hitting the market next year. TSMC is still using FinFET (the transistor design approach they first started using back in 16FF) here.
For Samsung, 3GAE supposedly started production this week. 3GAP is scheduled for next year. Here, they're attempting Gate-All-Around, which is the next major shift in transistor design.
Obviously, products aren't out yet. But, Samsung's latest promised gains for 3GAE and 3GAP are... not encouraging. The short of it is, if N3 and 3GAP deliver exactly according to the promises of today, I'd expect TSMC to stay winners.

2nm:
6 generations removed from the start. Too far off in the future for Drake, but hell, might as well finish things.
For TSMC, N2 is where they're attempting the transition from FinFET to Gate-All-Around. It is expected to start production in late 2025, so it probably means product in 2026.
As of last October, Samsung claimed that 2GAP will start mass production in 2025.

 

[Dakhil]

Speaking of NateDrake, here's my rough summary of the 13 October 2021 episode of Nate the Hate below since NateDrake mentioned the information from that episode as of today is still accurate.

• NateDrake believes Zynga's statement about not having a 4K devkit from Nintendo doesn't mean that Zynga didn't receive a 4K devkit from one of Zynga's publishing partners, who could have received a 4K devkit from Nintendo. Bigger publishing companies generally hire smaller companies as subcontractors and do send smaller companies devkits to work on games for certain platforms. But there's a possibility Zynga denied having a 4K devkit from Nintendo due to NDAs.
• NateDrake thinks Nintendo's technically not lying to investors when saying Nintendo's not supplying tools for developing games for a Nintendo Switch model with 4K support, but Nintendo's also not telling the entire truth, especially since Nintendo won't simply call the model the Nintendo Switch, but rather add a moniker next to the Nintendo Switch name (e.g. Nintendo Switch 2, Nintendo Switch Pro, etc.).
• NateDrake thinks Bloomberg was smart to obtain permission from a source in Zygna to name Zygna as the company that receive a 4K devkit, alongside mentioning that Bloomberg contacting 10 other third party developer companies, since Nintendo wouldn't be able to easily say Bloomberg's information is inaccurate.
• NateDrake thinks that part of the denial from Nintendo comes down to the branding of the model.
• NateDrake has heard from developer sources that the model's positioned as a revision, similar to the Game Boy Color and the New Nintendo 3DS.
• NateDrake thinks the name Nintendo chooses for the model depends on if Nintendo wants the count the model as part of the Nintendo Switch family or as a separate platform when talking about hardware sales.
• NateDrake believes that Nintendo is likely to have or will pressure Zynga to do an internal investigation, as well as Nintendo doing its own investigation, who's the source in Zynga who provided information to Bloomberg, which could damage Nintendo's relationship with Zynga.
• MVG agrees with SciresM that backward compatibility with Nintendo Switch games is not possible with the Nintendo Switch 4K, assuming that the Nintendo Switch 4K uses a GPU not based on the Maxwell architecture, mentioning that every Nintendo Switch game contains custom versions of the Maxwell GPU driver embedded in the game, with all the shaders required pre-compiled, in one package. MVG also mentions that developers can't simply take that package and compile it on a GPU not based on the Maxwell architecture. Instead, developers would need to recompile every game and provide a patch, or not offer backwards compatibility at all.
• MVG believes that the first possible solution is to provide patches for every game.
• MVG thinks the second possible solution is to open up a specific tool for third party developers that streamlines the update process that allow developers to take the game package and repackage it as a native game package for the new SoC.
• MVG believes the third possible solution is to add a Tegra X1 to the motherboard, citing the Nintendo Wii, the Nintendo 3DS, etc., as examples.
• And MVG believes the fourth possible solution is that backwards compatibility is not offered at all, where Nintendo brands the Nintendo Switch 4K straight up as a next-gen console, and Nintendo wants third party developers to jump on board, although MVG thinks it seems far fetched that Nintendo would do so.
• NateDrake believes that there's no way Nintendo won't provide backwards compatibility with Nintendo Switch games since it would send a message to consumers to not invest in digital games since Nintendo won't support consumers in the future.
• NateDrake doesn't deny the possibility that Nintendo could add the Tegra X1 to the Nintendo Switch 4K's motherboard to achieve 100% backwards compatibility.
• NateDrake also believes that there's a possibility Nintendo could be talking to Nvidia when designing Dane to add Maxwell GPU driver support to Dane, which could possibly achieve 99.9% backwards compatibility support.
• MVG said that the second possibility that NateDrake mentioned in terms of how Nintendo could achieve backwards compatibility with the Nintendo Switch 4K is possible.
• NateDrake believes that not offering backwards compatibility with Nintendo Switch games would cause Nintendo to lose a large amount of consumers since there's only so much bad business practices consumers can tolerate from Nintendo; and not offering backwards compatibility with Nintendo Switch games would be seen as one of the biggest anti-consumer moves.
• NateDrake thinks there's a possibility that the Game Cards for the Nintendo Switch 4K could very well be the same as the Game Cards for the Nintendo Switch, with the highest capacity staying at 32 GB. NateDrake also thinks that the Game Cards for the Nintendo Switch 4K could be slightly different, physically, to the Game Cards for the Nintendo Switch, like with the Game Cards for New Nintendo 3DS exclusive games, with the highest capacity possibly being 64 GB.
• NateDrake thinks Nintendo would announce the Nintendo Switch 4K six months before release. NateDrake also thinks that Nintendo could possibly announce the Nintendo Switch 4K on July 2022 with a release on October 2022, like with the OLED model, but at a risk at angering consumers who bought the OLED model, which NateDrake mentioned Nintendo has done before with Nintendo's previous products.
• NateDrake has heard from developer sources that development for games for the Nintendo Switch 4K are being targeted for completion on late 2022.
• MVG thinks that the Nintendo Switch 4K is more likely to be realistically released on early 2023.
• MVG think that Nintendo's using a DisplayPort 1.4 to HDMI 2.0b converter chip for the OLED model's dock due to economics, since the Mobility DisplayPort 1.2a to HDMI 1.4a converter chips used on the Nintendo Switch dock, as well as the HDMI 1.4 cables, are becoming harder to source.
• MVG's disappointed with the transfer speeds offered by the LAN port on the OLED model's dock.
• NateDrake will no longer refer the model as the Nintendo Switch Pro, but rather as the Nintendo Switch 4K, since Nintendo's releasing new Nintendo Switch hardware, and it has 4K compatibility, which will be achieved with DLSS.
• NateDrake doesn't know if the Nintendo Switch 4K will be marketed as a mid-gen refresh or a successor.
• NateDrake has heard from developer sources that the release window for the Nintendo Switch 4K is targeted at late 2022 to early 2023.
• NateDrake has heard a substantial amount of big third party developers received devkits in late 2020, and smaller third party developers received devkits on June 2021.
• NateDrake has heard from developer sources that there are games that are exclusive to the Nintendo Switch 4K, and won't be released for the Nintendo Switch (and the Nintendo Switch Lite).
• NateDrake has heard that developers are excited about the Nintendo Switch 4K.
• NateDrake has also heard developers were confused when the OLED model was released since Nintendo didn't send out new devkits for the OLED model.

Edit: Thank you @Raccoon for mentioning I could add a threadmark.

 

[ezidro3]

Here are some of his articles with explicit mentions of new models of the Switch, along with (what I think are) relevant quotes.

October 4, 2018 - Nintendo Plans New Version of Switch Next Year

Nintendo Co. plans to release a new version of its Switch videogame console next year to maintain the sales momentum of the device, according to suppliers and others with direct knowledge of the plan.

March 25, 2019 - Nintendo to Launch Two New Switch Models

One version will have enhanced features targeted at avid videogamers, although it won’t be as powerful as Sony Corp.’s PlayStation 4 Pro or Microsoft Corp.’s Xbox One X, according to parts suppliers and software developers for Nintendo who have access to a prototype of the machine. The other version is a cheaper option for casual gamers that Nintendo sees as a successor to its aging hand-held 3DS device, the suppliers and developers say.

June 12, 2019 - Nintendo Moves Some Switch Production Out of China, Adapting to Tariff Threat

The Wall Street Journal reported in March that Nintendo planned to update the Switch this year with two new models. One is set to look similar to the current model with beefed-up components, while the other is expected to be a less-expensive model with a new look.

August 24, 2020 - Nintendo Plans Upgraded Switch Console and Major Games for 2021

Nintendo Co. plans to debut an upgraded model of its Switch console next year along with a lineup of new games, people familiar with the matter said, ceding 2020’s holiday spotlight to rival devices from Sony Corp. and Microsoft Corp.

March 3, 2021 - Nintendo Plans Switch Model With Bigger Samsung OLED Display

Nintendo decided to go with rigid OLED panels for the new model, the people said, a cheaper but less flexible alternative to the type commonly used for high-end smartphones. The latest model will also come with 4K ultra-high definition graphics when paired with TVs, they said. That could intensify a longstanding complaint of developers, who have struggled with the difference in resolution between handheld and TV modes and now face a bigger gap between the two.

March 23, 2021 - Nintendo to Use Faster Nvidia Chips in New 2021 Switch Model

The new Switch iteration will support Nvidia’s Deep Learning Super Sampling, or DLSS, a novel rendering technology that uses artificial intelligence to deliver higher-fidelity graphics more efficiently. That will allow the console, which is also set for an OLED display upgrade, to reproduce game visuals at 4K quality when plugged into a TV, said the people, who asked not to be identified because the plan is not public.

May 26, 2021 - Nintendo Plans Upgraded Switch Replacement as Soon as September

The Switch maker’s response will be to upgrade its flagship console with a 7-inch Samsung Display Co. OLED display and faster Nvidia Corp. graphics silicon making it capable of 4K output when docked to a TV, Bloomberg News previously reported.

September 29, 2021 - Developers Are Making Games for a Nintendo 4K Console That Doesn’t Exist

Employees at 11 game companies said their teams were in possession of Nintendo’s 4K development kit for the Switch. The companies span the globe, ranging from large publishers to small studios and include at least one that has never made a console game before, Zynga Inc., according to the employees, who asked not to be identified because they weren’t authorized to discuss their projects publicly.

I think the takeaway here is that the 2019/Mariko Switch was mostly reported on accurately, with the upgraded components technically being true, albeit not utilized at all apart from battery savings. Lite was also practically spot on. It's the 2021/OLED model where information on OLED and what we know of Drake were conflated.

 

[Deleted member 887]

updated December 5 2022

Hey OldPuck, it's me again.
Oh hey, Anonymous Lurker, how is it going?

I noticed the thread got pinned, and thought I'd take a peek.
There is a lot of information in the OP if you wanna catch up.

Oh, I can't do that. There is just... so much info. Also, I'm not super technical, so even if I did read all of it, it would go over my head. Could you just tell me what we know.
Um... I honestly probably can't do that.

So you don't actually know anything?!?
We know more about this console than maybe any other unreleased console, ever.

Why can't you tell me anything then?
Imagine that you had the diagrams for a brand new car engine. You take them to an engineer, and excitedly ask questions about the car - how fast will it go? "Well, it depends on what sort of chasis they use..." what's the MPG like? "well, depends on how aerodynamic the body is and if they rate it for higher or lower RPMs" Don't you even know what the stereo system will be like??? "uh, absolutely not..."

It's the same here. We know a lot about the chip that the Next Switch will use, but there are also a lot of blanks. And we have good guesses for those blanks! But you don't care about that, you care about the games right?

Yeah, basically.
And a lot of smart people have made a lot of smart discoveries about the thing, but if I start saying specifics, I guarantee less than 1/3 of the regulars will agree with my final conclusions. At least 1 thing will be controversial. And that's why there is still a thread going even when there isn't new info.

It's like trying to solve a crossword puzzle when some of the clues are missing or blurry. We've actually filled a lot of it in, but the rest are reasonable guesses, or guesses based on guesses, or guesses based on those. It gets real tenuous after a while.

Okay but can't you tell me anything?
I'll tell you what I'll do. I'm going to compile the current as ofstate of affairs below. I'm going to summarize the best guesses, and I'm going to organize it from stuff we're 90% confident in all the way to "Guesses upon guesses upon guesses".

But I'm going to hide it all, because there has been a trend of disreputable people making bad-faith content out of the speculation here, and that causes lots of problems. I'm also going to hide chunks behind spoiler tags. I'm also going to update things here in the inevitable backlash as my fellow thread-dwellers correct me, but I'm not going to update this post as new data comes in.

Update: I said I wouldn't update this, but right after writing it, the Linux drops came out. In December, @JoshuaJSlone convinced me to make the post public, so I've brought it all up to date as best I can.

Got it?

Got it.

Spoiler: The NVN2 Hack/Linux drivers - the stuff we're pretty dang certain about

Earlier this year, Nvidia was hacked. They confirmed the hack was legit, and contained in the hack was the source code for what is essentially the graphics driver for a new Nintendo console, which tells us some stuff about the nature of the hardware in pretty clear terms. It's possible that the design has evolved since the hack, but unlikely it has changed considerably just due to the nature of chip development, and the state of the driver.

The new chip is codename "Drake" or T239. It uses the Ampere architecture the same as in Nvidia's RTX 30 series cards. It supports Ray Tracing and DLSS 2.0. It has 12 streaming multiprocessors.

Just a few days ago, Nvidia pushed a Linux kernel update that included references to T239, telling us that it has an octo-core CPU and is related to, but different from, Orin which is Nvidia's chip for Self Driving Cars, and in the same Tegra line as the TX1 chip that the current Switch uses.

Later, in progress Linux code for Drake was discovered online. Analysis (by me) confirmed that the driver architecture is very similar to Orin, with the self driving car elements removed. @Thraktor found Drake added a new feature Orin doesn't have - a File Decompression Engine for video games, likely to decompress textures directly to memory, a performance tool that other consoles have used off and on.

Spoiler: How do you know NVN2 is for Nintendo? How do you know Drake is for Nintendo?

NVN means Nvidia/Nintendo, and is the name of the Switch API. NVN2 is definitely for Nintendo. Drake is a chip that doesn't have another place in Nvidia's product line, is referenced in the NVN2 code, and is clearly built for a game console. If you want to dive further into that look at these posts

NVN/NVN2 Summaries
T239 Timeline

Spoiler: How powerful is all that?

Okay, here is the deal. We basically know how big the design is and how modern the design is, but we don't know how fast the clock speeds will be. Nintendo is going to be limited by battery life and heat in a way that other console makers aren't, and this thread has broad opinions on how far they can and will push it.

BUT - if we just plug in the Original Switch clock speeds into this design, you get something that looks a lot like an Xbox One but with ray tracing and DLSS.

This level of power will make games like The Witcher III - the "impossible ports" of the current Switch era - into "straightforward ports."

DLSS lets those games get to 4k, like the Pro consoles of the last gen, with a lot less power.

RT opens up new options for cross-gen ports, and future "impossible" ports.

First party games have a bigger bump than from Wii U to Switch.

Also: this is only on GPU level. The CPU/Memory situation is very similar - we know the outline, but the details are unset, everything points to a sort of "8th gen plus" level of performance.

Spoiler: Octo-core?

8 CPUs. The current Switch is 4 CPUs. The Xbox One, Xbox Series S|X, the Playstation 4 and Playstation 5 all have 8 CPUs.

The last gen Xbox/Playstation CPU was notoriously weak. The current Switch CPU is a Cortex A57, which already performs similarly to the Jaguar CPUs in the Xbox One and the Playstation 4 - it just has half as many. Even if Nintendo doesn't raise the clock speed, 8 CPUs would all but guarantee that the CPU performance of Drake would outrun the last gen.

Spoiler: What in the hell is a Streaming Multiprocessor?

It's a way NVidia organizes the various bits and bobs in the GPU. For comparison, AMD uses a similar structure but calls them "compute units." The Xbox One had 12 of these, same as Drake. But Drake is using Ampere, a much more modern architecture than the early RDNA version on the last gen consoles.

Imagine a GPU's power being like a river. The amount of water that a river can move is based on how big it is and how fast the water moves. 12 SMs tells us how big the GPU is, but the clock speeds tell us how fast. And right now, clock speeds are just guesses..

Spoiler: RAY TRACING NO WAY

Yes way, but don't get crazy with it. Remember what I said about clock speeds? The RT cores in Ampere are significantly better than the ones in PS5/Xbox Series consoles, but are going to run much slower, just for battery life/heat reasons. Nintendo is probably going to run the GPU clock at speeds beneath any other Ampere graphics card on the market, at least in handheld mode.

This makes it really hard to know what it can do, and it means that Ray Tracing based ports will likely have to "cut down" the ray tracing to fit. BUT - they won't need to build a whole new lighting solution from scratch. This is one of the reasons we say that "new impossible ports will be possible."

Spoiler: Explain DLSS like I'm 5, why is it a big deal?

Higher resolution = higher detail, but more pixels to draw, so slower
Lower resolution = lower detail, fewer pixels to draw, better frame rates.

Temporal reconstruction is a technique that lets you break this relationship, so that you can get most of the detail of high resolution, with most of the frame rate of lower resolution.

Most modern AAA games use a form of temporal reconstruction, DLSS is a specialized version that uses AI techniques and is hardware accelerated for quality and speed that generally outperforms other solutions.

If you have heard of "checkerboard rendering" from Playstation 4 Pro, that is temporal reconstruction. Checkerboarding makes a 4k image out of 2k pixels. DLSS can make a 4k image that looks as good or better out of just 1k pixels. The Pro consoles last gen needed a huge amount of excess power just to get to 2k, and then use checkerboard rendering to get to 4k. DLSS gets you that Pro console experience with a lot less GPU power.

Update: Nvidia has since announced DLSS 3.0, which uses upgraded hardware that isn't in Ampere thus likely not in Drake. There is no indication that DLSS 3.0 will be possible on Drake. DLSS 3.0 isn't designed for 30-60fps games, and because of that, is probably not a great fit for Nintendo anyway.

Spoiler: Trusted sources - the stuff we feel pretty good about

This is the stuff that journalists with good track records say. Plans can shift, journalist's sources can have incomplete pictures, but these are things stated from folks who have a record and aren't making stuff up, and that at least multiple sources have said.

It's a Switch - in other words this is not a new TV only console, or a VR headset or anything like that, but a device that Switches, and plays Switch games. It is slated for release in the first half of 2023. It will have at least one exclusive.

Spoiler: Ya'll are moving the goal posts on the release date since 2020, why should I trust you?

We're a bunch of forum goers and internet randos, so trust is maybe a little out of the question

The release date issue has been thrashed to death and summarizing it here would double the length of this already gigantic post. Short version: For over a year now we've known that 3rd parties were expecting a release sometime in the H2 2022 - H1 2023. Some optimists consistently want to see it in the earliest possible part of that window, while more aggressive pessimists have disbelieved much of what you're already reading here. That causes a lot of "boom and bust" cycles on the release date.

Spoiler: What's it called? Is it a successor or a revision?

No one knows. It's definitely what you would expect from a successor in terms of an internal design, but how Nintendo will position it, how 3rd parties will respond, and how long OG Switch support will continue are all open questions.

Lots of folks have theories many of them well supported. But no one knows, and this is one of the many things the thread discusses.

Spoiler: What are the reasonable guesses - ie, stuff we have no direct source for but Make Sense based on what we know

This thread knows a lot about silicon and game dev, and has done a lot of research. If you make some reasonable assumptions, you can make some decent guesses, but they are just that, guesses.

The base assumptions here are this device basically shares the current Switch's form factor. Nintendo will take advantage of the mobile market and purchase off the shelf components where reasonable to do so, and Drake, the new SOC, shares some architectural similarities to Orin, NVidia's platform for self-driving cars.

Spoiler: Memory

8MB of RAM with good throughput are reasonable baselines.

Nintendo will likely use LPDDR5 RAM, as that's pretty standard for mobile hardware manufactured now, and matches the RAM in Orin. Nintendo will likely use off the shelf memory, because with the number of mobile devices out there, Nintendo can get high quality RAM for cheap. The NVN2 hack doesn't list the amount of memory, but it does imply a memory bus width of 128 bits - half of Orin's.

Given all this, the minimum RAM is 8GB, twice the current amount in the Switch. Nintendo doesn't skimp on RAM, and the market actually makes 12GB a potentially cheaper alternative, though it would eat more power/space and generate more heat. As such 12GB has sort of become the standard assumption.

Spoiler: GPU clock speeds

TL;DR: Expect clocks to stay the same, or to have a modest bump, and most of the new power to come from the large number of SMs and cores

We can look to Orin for reasonable Max and Minimum GPU clock speeds. Orin doesn't go below 420Mhz in its most power saving mode, that is likely where Ampere's efficiency bottoms out. And even in Max power mode, the GPU clock doesn't go past 1.3Ghz, but requires active cooling that might be unreasonable for the Switch even in docked mode.

@Look over there has determined that Drake might be limited by memory bandwidth if clocks go past 1GHz. @Thraktor sees that even at the 420Mhz clock, the GPU eats more power than the OG Switch, meaning Nintendo will need to find savings elsewhere to keep battery life up.

The current Switch is 768Mhz in docked mode, 300Mhz by default in handheld mode, but many games use a 460Mhz mode in handheld. 420Mhz would be the bottom of the range, 460Mhz would be enough for backwards compatibility in handheld, unless Nintendo starts to get truly desperate on the battery life front.

DLSS creates some advantages for games to have very close to 2x GPU in docked mode. So 840-920Mhz seems like a reasonable target, assuming heat isn't a problem.

Spoiler: FLOPS?

FLOPS are not a great measure of GPU power when it comes to games. But.

CUDA cores do 2 FLOPs a cycle, 12SMs=1536 CUDA cores, FLOPS=2*1536*Clockspeed

Drake at current Switch clocks = 2*1536*(768 or 460) = 2.3 TFLOPS docked, 1.4 TFLOPS handheld.

Spoiler: CPU?

8 powerful CPU cores, running at 1-1.25 Ghz

The Linux driver updates for Drake say that it runs 8 ARM cores in 1 cluster. There are only a three CPUs that can do that.

A715, which is very new, and doesn't offer 32 bit support, which Switch probably needs.
A710, which is also fairly new, and doesn't have great power/performance characteristics
A78C, which is a variant of the same cores in Orin, and is designed specifically for gaming. This is almost definitely the CPU core in Drake.

A78C is a nice upgrade over the A57 in the current Switch. Benchmarks tend to show similar performance to Zen 2 (the chip in PS5 and Series S|X) at similar clock speeds. Orin power data suggests that an 8 core A78C cluster uses about the same about of power as the current Switch's 4 core A57 cluster, at the same clocks.

Considering how hungry the GPU is, and how powerful the CPU is, keeping the CPU at roughly the same clocks as the current Switch makes sense. A modest bump would be very cheap, if there was some headroom.

Spoiler: Screen

There are no indications about the screen, but as OLED screens are more common and thus driving down the price and Nintendo is unlikely to want a downgrade from their current “premium” model, and OLED screen seems likely.

720p is a sweet spot at the screen size of the existing Switch, but if Nintendo goes with faster handheld clocks, and phones drive OLED production, a 1080p screen is certainly possible.

Spoiler: “Process node”

This is one of the more controversial questions, historically. Smaller node = more expensive device, but better power efficiency

Orin is on Samsung 8nm, as is the entire Ampere desktop GPU line. This should be considered the default. However, there has been considerable discussion on whether the leaked Drake design is too big to run on Samsung 8nm and have decent battery life.

Orin's power data shows that Drake would almost definitely draw more power than the original Switch, even at pretty low clocks, but not so much more power that Samsung 8nm is impossible.

Spoiler: Let's go hog wild. Give me the wild guesses from the Dark Web

These are unverified wild scribblings, but not stuff that has been clearly disconfirmed

Spoiler: Chinese language forums

There are some plausible "leakers" on Chinese language forums, which we have collectively followed, and it turns out they follow us. There are some indications that production is moving out of the prototype stage, but even if these leakers are legitimate factory employees, they've a record of not fully understanding what they're working on and "leaking" things that were not new.

Some of this seems to indicate a possible change to the kickstand design so instead of being a solid piece like the OLED model, it runs the outline of the device, ala some HP devices on the market.

Spoiler: Famiboard Users

A certain user here has claimed a connection at Rockstar in Edinburgh. They've said several things about the device are consistent with more vetted statements, but their distinct claims are RDR2 is coming to Drake and that the device is 12GB

They've also claimed the devkits were earlier 6GB, a claim which has disputed plausibility, and that partners were recently briefed, which has since been confirmed by Nate.

Spoiler: The Spanish Leak, The Other Chinese leak, and more

There are a series of less plausible leaks. They're not included here, even in the wild speculation section, because you need some standards. Two notable ones I include here to debunk.

One was a Chinese forum post that listed very plausible specs for a device that seemed to match a lot of what we know. Some detective work found that the almost definite origin of this "leak" was a speculative post of mine on this very board, as it contained an obvious mistake I made despite rewording.

The second was an extensive and impressive Spanish insider. There was lots of potentially interesting new information, specifically about power management on the device moving towards dynamic clocks rather than specific profiles. However, this "leak" said several things about the CPU that were directly contradicted by the Linux driver updates. It seems that this person has been posting 'Switch 2' updates for some time, incorporating more information from the community, but in order to keep their story consistent, has had to hang onto some incorrect prior guesses.

Spoiler: I've still got questions

Spoiler: I don't believe Nintendo will ever release a modern device!

Then I'm not sure this thread is for you, but if you don't believe the NVN2 hack you might be in denial. The Linux updates aren't even hacks, they're public dumps of technical data from NVidia. The above is very conservative.

But if it helps, this is consistent with Nintendo's trend of staying behind their competitors in power - it's just that Nintendo has had a "catchup" generation because the Wii U era was short.

Spoiler: Price?

Dunno. Drake/T239 is a high end chip, however the fact that Linux drivers have started to drop implies that NVidia plans to use the chip in non-Switch devices. This likely reduces costs on Nintendo’s side, as it means NVidia probably fronted some of the dev costs.

Spoiler: New Gimmicks

There is no indication of any. Nvidia's tech is oriented at AI and real time imaging, but a lot of that hardware is stripped away in Drake or used for DLSS.

There have been discussions about:

• More cameras
• Dropping the IR camera
• Handheld streaming to the TV

But these are just idle speculation. There is no evidence.

Spoiler: Size?

No clue, but we're running with "roughly the switch's shape, so that Joy-Cons, Docks, coves and alternate controllers are compatible" but that's just an assumption

Spoiler: battery life?

Will depend on myriad factors, but we're all roughly assuming that the increased power of the device will be countered by slightly improved battery efficiency to land right back in "Day 1 switch" territory

Spoiler: Name?

Dunno, again, this is about how Nintendo choses to position the device as much as anything else.

My personal vote is for "Nintendo Swuntch"

Spoiler: Release date?

H1 2023 seems Highly Likely - that period has been tagged back as the date devs were expecting to release games in the earliest reporting, and tracks with other info ("partners were recently briefed). Most folks expect before Tears of the Kingdom, or a simultaneous launch, but it's an open question.

Spoiler: Launch Title?

No clue, again it is unclear if Nintendo will launch it with exclusives, or just have "plays better on Drake" style updates to existing titles

Spoiler: Boost mode?

Most folk expect that at least some games will get some kind of "boost mode" where they can more consistently reach their frame rate and max res. Big games might get some Drake specific patches. Again, this comes back to how Nintendo markets the device.

Spoiler: Okay actually I lied, I'm super technical, can you just list what we know without explanation

Uhhh, mostly? This will include some "99% sure, but could technically be wrong" guesses in here, but...

CPU: 8 Core ARM Cortex-A78C, in a single cluster. 32k L1 data cache, 32k L1 instruction cache, 256k L2 cache, 4M of L3 cache

• the A78C supports a config that double these cache numbers
• Due to the nature of the A78C power curve, a ~1-125GHz clock seems likely (<0.5W per core on 8nm)
• A 1.75-2Ghz clock represents the "realm of possibility" maximum (<0.5W per core on N5)

GPU: Nvidia Ampere with 1536 CUDA cores (12 SMs/6 TPCs)

• Original Switch Clocks seem likely, 460MHz/768Mhz, for 1.4/2.3 TFLOPS
• 1Ghz in docked is the likely maximum.
• The ratio between docked clocks and handheld clocks is not in any way set in stone, but ~2x matches with DLSS needs and screen size quite nicely.

Memory: 2x 4GB 64-bit LPDDR5 modules @34 GB/s

• This is the minimum Orin config, and matches with the 128-bit bus that is available in T239
• 2x 6GB 64-bit LPDDR5 @102 GB/s seems extremely likely based on the availability of those modules
• Orin runs a 3200MHz memory clock in all but the strangest power configurations
• But a 2100MHz clock is where Orin Nano lives, and might represent a source of power savings.

Update: Clarification from @Look over there, @Alovon11, and @Brofield. Updated on the 20th with info from @Cybergatuno because not doing so seemed dumb, even if it came in after the 18th. Added some info I found in the Linux drops from Nvidia, and updated with @BlackTangMaster's power data and @Thraktor's power analysis.

 

[LiC]

Here's the final post I'll make on the tiresome "how do we know this isn't all unrelated to Nintendo" subject. In the future let's just refer people back to this instead of rehashing the same arguments over and over again.

I will start by saying that even though the "NVN might not be for Nintendo" etc. arguments are 100% wrong, we can't say for certain what will or won't happen with specific hardware Nintendo releases in the future, because we don't have a crystal ball. With that facile point out of the way, let's work through three questions here.

---

Q1: Is NVN2 being created for Nintendo?
A1: Yes.

• NVN is a custom graphics API available only in the Nintendo Switch SDK. It has never been shown to exist in any other form.
• NVN can run on Windows, but only to aid in the development of Switch games.

Reference implementation on non-NX platforms: Provide a reference implementation that can be used on Microsoft Windows with NVIDIA GPUs and behaves as closely as possible to the NX implementation. This allows application code to be developed and debugged on Windows with only minimal differences from the native NX device that might require \#ifdefs or alternate code paths.

• NVN's leaked source code is full of references to NX (Switch), HOS (the Switch's Horizon OS), and Hovi (a codename for Nintendo). Sections that were newly added to NVN2 still have such references.

NVN2 -

This will need to change on HOS. On HOS, the application is responsible for providing the scratch memory, so that will need to be factored in somehow.

On l4t we don't have memory requirements but we will for HOS

Hovi wants to be able to present from a different thread as the window creation.

---

Q2: Is NVN2 intended to run on T239?
A2: Yes.

• Where NVN's source code referenced hardware values from Tegra X1's GPU, NVN2 now references values from T239's GPU, GA10F.

NVN1 -

// Number of warps per SM on TX1 hardware
#define __NVN_NUM_WARPS_PER_SM_TX1                  128

// Number of SMs on TX1 hardware
#define __NVN_NUM_SMS_TX1                           2

NVN2 -

// Number of warps per SM on ga10f
#define __NVN_NUM_WARPS_PER_SM_GA10F                48

// Number of SMs on on ga10f
#define __NVN_NUM_SMS_GA10F                         12

• The NVN documentation has a section about which GPUs are supported on Windows, and which of those should be used for maximum compatibility with the NX hardware. This originally stated that second-generation Maxwell GPUs were preferred (since the Tegra X1 has one), but NVN2's documentation has changed this to state that Ampere GPUs are preferred (and GA10F is an Ampere GPU).

NVN1 -

GPU: Any NVIDIA GPU from the Maxwell GPU families supported by the custom driver build. A second-generation Maxwell GPU (e.g., GeForce GTX 960, 970, 980) is recommended, because it provides nearly all the NVN API functionality supported by the NX processor.

Second-Generation Maxwell GPUs: Second-generation Maxwell GPUs have nearly the same feature set as the GPU core inside the NX device. [...] This class of GPUs is strongly recommended for NVN development on Windows.

NVN2 -

GPU: Any NVIDIA GPU from Turing GPU family onwards supported by the custom driver build. Ampere based GPU's are preferred since they are more compatible with NX.

Turing GPUs: The NVN Windows reference implementation is supported on Turing-based GPUs but Ampere-based GPUs are more preferable since they are more compatible with the NX implementation.

Ampere GPUs: Ampere GPUs are the compatible with the NX product.

---

Q3: Is NVN2 going to be used for the currently rumored new Switch model?
A3: We can't know this for certain with the information we have, but consider the alternative: A new Switch model is rumored to use a more powerful Nvidia chip to support DLSS and ray tracing, then a data breach reveals a new version of the Switch's graphics API which runs on a more powerful Nvidia chip and supports DLSS and ray tracing on an "NX product," but it's not for the Switch model that was rumored.

• 

NVN2 -

Deep Learning Super Sampling
Overview

Deep Learning Super Sampling (DLSS) is a technology that enables high-quality upscaling of rendered images to a higher resolution by leveraging a pre-trained deep neural network that operates on lower-resolution input images.

This document focuses on aspects of the interface specific to NVN. [...]

Ray Tracing

The NVN ray tracing API is similar to the Direct3D and Vulkan ray tracing APIs. This section provides a high-level overview of the API. It assumes some familiarity with the basics of ray tracing.

 

[Dakhil]

• In 2021, Nate the Hate reported there were rumblings within the development history of Nintendo distributing devkits for new hardware with 4K support achieved with DLSS
• Bloomberg later reported hearing the same, with some of Bloomberg's reports being corroborated by Nikkei
• In mid 2021, Bloomberg reported the release of new hardware was imminent, which turned out to be a false trail, with the OLED model being released
• But in autumn 2021, Bloomberg doubled on reports on there exists devkits for new hardware with 4K support, citing 12 developer contacts having access to it, with a release timeline of late 2022
• A month later, NateDrake independently corroborated Bloomberg's report in autumn 2021, but with a slightly different release time of late 2022 - early 2023
• In 2022, there's an illegal leak of Nvidia files, which has mentioned NVN2, the successor to NVN
• MVG has heard some developers at GDC 2022 mention working with updated Nintendo Switch devkits, which MVG kind of assumed to be devkits for a next-gen Nintendo Switch (MVG mentioned developers never mentioned "Nintendo Switch Pro")
• NateDrake has heard his developer contacts mention having new devkits for a new Nintendo Switch, which is more powerful and has more features, and mentioned the new hardware seems more like a revision than a successor, but that depends on how Nintendo wants to brand it
• John Linneman thinks everyone used the "Nintendo Switch Pro" moniker too liberally since "Nintendo Switch Pro" was easy to say, although whether the new hardware was positioned as a mid-gen refresh or a successor was unknown
• NateDrake thinks Sony and Microsoft could have marketed the PlayStation 4 Pro and the Xbox One X as a successor if Sony and Microsoft wanted to since there were enough meaningful upgrades in terms of performance uplifts and features (e.g. 4K support)
• John Linneman mentioned the PlayStation 4 Pro and the Xbox One X released 3 years after the PlayStation 4 and the Xbox One
• MVG always felt the "Nintendo Switch Pro" is a mid-gen refresh still using the Tegra X1, but with increases in the memory bandwidth, memory size, and the CPU and GPU frequencies (basically the same hardware with more oomph), and is still binary compatible with the entire Nintendo Switch library without needing patches (MVG said this is his interpretation)
• When inquiring his developer contactors on summer 2022 about whether the new hardware is still on track to release on late 2022 - early 2023, NateDrake heard more rumblings the new hardware is no longer scheduled to be released in late 2022 or early 2023
• That surprised NateDrake, given what MVG has heard at GDC 2022
• NateDrake mentioned all prior reporting was correct for its time, but plans have changed
• John Linneman believes the new hardware described by Bloomberg no longer exists and that nobody should expect new hardware to be released in 2023 based on the timelines lining up, although he admits nobody's 100% certain
• John Linneman thinks Nintendo could announce new hardware in autumn 2023, with a demo event in early 2024, and with new hardware released in spring 2024, based on what Nintendo did with the Nintendo Switch prior to release
• John Linneman wonders of Metroid Prime 4 could be a launch title for the new hardware, with Tears of the Kingdom being too late, although Nintendo could release a Tears of the Kingdom port for the new hardware
• John Linneman and MVG believe the reason ADEV (the OLED model's devkit) exists is to allow developers to see the game on the OLED display whilst developing the game
• NateDrake mentioned ADEV is not conflated with the new devkits with 4K support since the OLED model was always meant to be its own separate thing
• NateDrake mentioned the illegal Nvidia leaks mentioned a SoC codenamed Drake that supported features, such as ray tracing and DLSS
• NateDrake and MVG mentioned having no first-hand confirmation on which SoC was used for the devkits with 4K support (MVG mentioned he kind of assumed the Tegra X1 was still being used)
• NateDrake wonders if because the SoC evolved and became more efficient over time, Nintendo issued a recall for the devkits Nintendo has been working on and has distributed, and distributed better devkits
• NateDrake also wonders if Nintendo pivoted on a certain element (not a concept) of the hardware itself, causing the new hardware to be better, therefore the new hardware is no longer planned to be released in 2023 (perhaps early - late 2024 hypothetically?)
• MVG thinks it's possible to assume Nintendo did initially send out devkits to third party studios, so the third party studios could become familiarised with the new hardware, whilst Nintendo figures out what kind of battery life should the new hardware have; once Nintendo got metric from third party studios on the average battery life, Nintendo asked third party studios to send back the devkits to initiate phase 2
• John Linneman mentions this is what other manufacturers (e.g. Microsoft, Sony) do as well
• NateDrake reiterated the previously reported information was true for its time, but was ultimately incorrect due to being reported on too early
• NateDrake thinks Drake, which was mentioned in the NVN2 files, could still be developed by Nvidia, still be tested, and be repurposed for what Nintendo planned for what's coming next
• MVG mentions that NVN can run on a multitude of Nvidia GPUs despite being geared for the Nintendo Switch, and has libraries that can be compiled on different OSes (e.g. Microsoft Windows, etc.) depending on which Nvidia GPU was used; and MVG expects no different for NVN2
• John Linneman thinks Nintendo will be fine without releasing new hardware into 2024 since the Nintendo Switch is the best selling console of all time, and there's a huge audience base that still wants to buy games for the Nintendo Switch, with the only problem he foresees is third party Nintendo Switch ports
• John Linneman believes the strength of Nintendo's first party games, as well as indie games, could still make the Nintendo Switch an appealing product into 2024
• MVG plays a little bit of a devil's advocate by mentioning there are first party (e.g. Age of Calamity, Pokémon, etc.) and third party games that have performance issues on the Nintendo Switch
• MVG expects Tears of the Kingdom to be a generational leaps in terms of visual fidelity, compared to Breath of the Wild
• NateDrake mentioned that in an episode of Nate the Hate on September 2022 that John Linneman believes generational leaps in terms of visual fidelity are a thing in the past, but not because of hardware limitations, but rather because developmental tools have become very advanced
• John Linneman mentioned developers were experimenting with different development tools and techniques during the Xbox 360 and PlayStation 3 era, during which best practices haven't been established yet
• MVG mentioned the biggest thing in rendering techniques currently is that developers are pushing as many GPU instructions as possible to RAM, and pushing it in one go, due to the abundance of memory and memory bandwidth on consoles
• MVG mentioned there haven't been many third party Nintendo Switch ports due to the Nintendo Switch having a narrow memory pipeline
• Although the Nintendo Switch's hardware is outdated, John Linneman still thinks the fact the Nintendo Switch can still run nice looking games at 60 fps whilst running at less than 10 W (e.g. Burnout Paradise) is still somewhat magical
• John Linneman thinks third party developers could re-explore some of the Xbox 360 and PlayStation 3 era games in their back catalogue, re-tool them, and sell them as remasters, re-releases, etc., on the Nintendo Switch, and get a decent amount of sales due to people's nostalgia of the Xbox 360 and PlayStation 3 era
• NateDrake thinks third party support for the Nintendo Switch will wane down when Nintendo introduces new hardware in 2023 or 2024 due to the impracticality of targeting the Nintendo Switch hardware when developing a port of AAA games, and third party developers are communicating to Nintendo they want to release ports of AAA games to Nintendo's platform, but they need more powerful hardware to do so
• NateDrake thinks that based on Drake, the features it has, and after making assumptions about the frequencies, etc., the new hardware could be comparable to the PlayStation 4 Pro, which is more than enough to receive downports of AAA PlayStation 5 and Xbox Series X|S games, and there's nothing stopping the new hardware from receiving ports of AAA games
• NateDrake believes Microsoft has done Nintendo a huge favour with the Xbox Series S, since if the new hardware can get Xbox Series S ports without too much compromises and can be upscaled using DLSS, Nintendo could be placed in an advantageous position with competitive hardware
• John Linneman iterates DLSS isn't magic and there's a computational cost associated with DLSS, and the Xbox Series X|S already has access to FidelityFX Super Resolution, which is comparable to DLSS, despite not being as good as DLSS
• MVG mentions DLSS introduces input lag, which John Linneman chimes in and says that's the case for DLSS 3, although John Linneman mentions games running at 120 fps using DLSS 3 does have lower latency compared to games running at 60 fps natively
• John Linneman mentions DLSS 3 isn't beneficial if a game targets 30 fps from an input lag perspective (although visual fidelity could still be ok with DLSS 3)
• John Linneman hopes Nintendo installs a VRR enabled display for the new hardware, although he thinks Nintendo's not likely to do so, with NateDrake chiming in saying he'd hope the display also supports HDR
• John Linneman thinks Nintendo could be afraid of OLED burn in, and HDR may require displaying at higher nits of brightness, which is why a display might not support HDR, although there's a possibility the OLED model supports HDR when displaying at lower nits of brightness
• NateDrake hopes that if the new hardware's display doesn't support HDR and VRR, the new hardware support those features in TV mode
• MVG thinks the new hardware is likely a "Nintendo Switch 2" since the Nintendo Switch has been so popular, although the possibility Nintendo does something unexpected can't be denied
• John Linneman mentioned Nintendo doing something unexpected was from a specific era (under Satoru Iwata's leadership), and with Nintendo's current leadership (with Shuntaro Furukawa) being very conservative, Nintendo could simply release a "Super Nintendo Switch"
• John Linneman thinks the reason for a dry spell of first party Nintendo Switch games is because Nintendo released most of the first party games way too early, and now Nintendo's on the drawing board for the new hardware
• Although MVG doesn't want to talk about in detail about backwards compatibility, MVG thinks backwards compatibility with the entire Nintendo Switch library could be one way to entice regular consumers to upgrade from the Nintendo Switch to the new hardware
• John Linneman thinks Super Smash Bros, Mario Kart, Splatoon, and the Warrios games (e.g. Hyrule Warriors, etc.) have run its course and reached a dead end with the Nintendo Switch (since the Warriors games have marketed graphical improvements), although Nintendo could follow up with Mario Kart 9; and he doesn't know what Nintendo's going to do next
• NateDrake thinks there needs to be more exciting, fresh, new innovations for game designs of some of Nintendo's iconic franchise (e.g. Splatoon)
• John Linneman thinks perhaps a hub world of Splatoon could be directly linked to all the matches happening within an instance, see all the matches happening, and jump directly to one of the matches, thanks to more hardware power
• John Linneman also thinks perhaps there's a single player level game with an instance of multiplayer running in the skybox below the player thanks to more hardware power
• NateDrake and John Linneman believe the next Super Smash Bros game needs a reboot, although they don't know how Nintendo could make it new and fresh without Masahiro Sakurai and without alienating the Super Smash Bros community
• NateDrake thinks new hardware will be released at around 20 months (around holiday 2024 - early 2025), with MVG concurring
• NateDrake mentions hardware is a very difficult, fluid thing, and the industry's constantly changing, which results in previously reported information that was accurate at its time ultimately being inaccurate, although that doesn't necessarily make previous reporting wrong
• MVG is ultimately glad Nintendo's moving forward with releasing a next-gen Nintendo Switch
• MVG thinks the next-gen Nintendo Switch is a hybrid model, although he admits he's being safe
• NateDrake thinks the next-gen Nintendo Switch will retain the portability aspect, but he doesn't expect a 1:1 match with what the Nintendo Switch is, and there's a differentiating factor between the next-gen hardware and the Nintendo Switch
• John Linneman thinks the next-gen Nintendo Switch is a continuation of the Nintendo Switch, but better and faster
• John Linneman believes the next-gen Nintendo Switch needs to support Unreal Engine 5 specific hardware features (e.g. Nanite, Lumen, virtual shadow maps, etc.), which he admits is a tall order since the Steam Deck struggles with maintaining performance when supporting those Unreal Engine 5 hardware specific features; and the next-gen Nintendo Switch needs to properly support 4K TVs (not necessarily render games at 4K)
• John Linneman isn't confident the new hardware will have ray tracing support in actuality (but the SoC having support for ray tracing is possible, although a different matter) since ray tracing drains battery life and won't be able to compete with the PlayStation 5 and the Xbox Series X|S, and RDNA 2 already isn't great with respect with ray tracing, with MVG concurring
• NateDrake thinks a limited form of ray tracing is possible, but only when in TV mode, with John Linneman and MVG saying Nintendo and third party developers don't want that to happen
• NateDrake thinks the reason why Nintendo won't support certain hardware features isn't because Nvidia isn't able to provide (e.g. ray tracing), but rather because certain hardware features don't fit with Nintendo's goals
• John Linneman wonders how Nintendo could have hypothetically handled a mid-gen refresh (for the Nintendo Switch family), with NateDrake mentioning that's for another podcast episode

 

[Paul_Subsonic]

Introducing the DRAKE DLSS ESTIMATOR 6000 :

Drake DLSS Estimator 6000 | CALCONIC_ Calculator

Custom Drake DLSS Estimator 6000 calculator, built using CALCONIC_

www.calconic.com

This new version has some massive changes from the last one, and the results are quite different.
Significantly more accurate (probably) and significantly more optimistic than the last one.
For 4K, the new calculator indicates usually around a 35% smaller time than the old one. Reasons for that detailed in the technical section below.

Spoiler: The technical stuff

Let's go quickly over the main changes before going into the details.

Patch notes :
-Complete replacement of all DF data by official Nvidia data
-less variables, which means a smaller margin of error
-Introduced 1080p, removed 1440p
-Linear function used previously replaced by a 3rd degree polynomial (most important)
-Making it look slightly cooler

First thing I want to talk about is a correction : I claimed in a "technical stuff" section in a previous post that DLSS was not just linked to output resolution; I was wrong. This assumption came from the DF data in their second video, but the data in that video turned out to be... meh. Main issue is that the methodology worked perfectly for the purpose of the video, which was comparing FSR to DLSS, but the data is just not good for our use case.

The old calculator was entirely based on that DF data, and that data has been completely phased out of the calculator. Now the calculator uses the data from the document I linked and said was "The One document to rule them all". First advantage is, as said previously, that less variables are involved, reducing the number of ways this can go wrong. But that's not really what's important. That document is from Nvidia, and has data for a lot more GPUs, although only Ampere has been used here for reasons I'll detail later. This means the data is more accurate, but the most important is that the bigger number of GPUs tested allows us to verify if the cost of dlss scales linearily with performance, as supposed until then.
It does not.
It is hard to come to a conclusion as the data is still not as much as I'd like, but using Excel, we can see the curves are akin to a polynomial. I have, as such, asked Excel to make polynomials out of the curves. They ended up being of the third degree, simply because there’s only 4 GPUs whose data I can use.
Here are the curves, with the X axis being the tensor performance and the Y axis the speed of the DLSS calculations (not how long it takes, but the speed; a higher value means faster so a lower ms count).



That change in the way scalability is calculated makes a MASSIVE difference to the results, and is the main factor of change compared to the old calculator.
Here is an illustration : the intersection is the 3080; X coordinate is tensor performance, Y coordinate is the speed of the DLSS calculations.
The orange curve is how I previously thought it would scale, blue curve is how it actually scales.




To give you an idea, Drake would be around 0.07 on the X axis. We see that this new method of calculation heavily benefits Drake.

Using this data, I discovered a lot of things. First of all, the scaling of DLSS speed depending on resolution. It scales kinda linearily, but very loosely. I’ll first talk about the 1080p to 4K difference. I have noticed that the more powerful a GPU is, the smaller the difference between 4K and 1080p speed is; and the less powerful a GPU is, the more that difference tends to be the same as the difference between the number of pixels (so 4x). For example, For Drake, the difference between 1080p speed and 4k speed is very close to 4x because Drake is very not powerful, but for a 3090 it's more like 2.9x.
4K to 8K is significantly weirder. It starts at around 4.3x, then goes up, then down. At first I thought this was some proof the Excel predictions were just not working, but after looking at the Nvidia data, we do see 4k to 8k going up then down.






It isn’t shown in this graph, but I also made 1080p to 1440p, which also looks very weird : starts at 2.2x difference, goes down to 1.5x then starts going up again. So 1080p to 4k looks normal and makes sense, but 4k to 8k and 1080p to 1440p are much weirder for not apparent reason. Btw, talking about 1440p…


I removed it because I’m not sure how accurate it is. Even just looking at the raw data through a graph, the 1440p speed as a function of tensor performance looks... wacky (cf first graph, the orange curve). It looks like the other curves but on steroids. And the fact that the 1080p to 1440p difference curve also looks wacky does not convince me of the validity of my results. I do think in the future I might add back 1440p; but I want to be more sure of what’s going on before, and I want to first put in place the enhancements I mention towards the end.

Now is the time to explain why I didn't use Turing to improve the prediction. I mean, I say now's the time but any point could be the time, it's not linked to any of the other points. In the document, 3 Turing GPUs are benched : 2060 super, 2080 laptop and 2080 TI. First problem : laptop GPUs have special power limiting software that makes it probable the laptop gpu was not running full speed. And we don't know what speed it's running at. This is also reflected in the graphs below, where we see the 2080 laptop is way closer to the 2060 super than it should be.





But that means for Turing we only have 2 GPUs left. How do you infer a curve out of that ? You don't really. That's not enough. 4 is already limiting, so 2? Don't even think about it. And that is why I didn't take into account Turing GPUs in the calculator.

I'll now talk about the sparsity improvement. You may recall that earlier, we used DF's data with a 2060 to estimate the improvement from sparsity. Turns out that improvement massively depends on the card : the faster the card, the smaller the difference. The resolution also plays a role, but a minor one. As such, at the 2060 performance level, we see a 25% improvement from sparsity. 25% is what was found with DF data with a 2060, so that was close. On the other hand, at the 2080 ti performance level, the improvement is merely 9%.


Now let’s talk limitations of my simulations.


First of all, because of the data itslelf. We are talking about official data from Nvidia, but they said they mesured that experimentally by running a command prompt. But anyone who has ever done physics knows that experimental measurements will always be imperfect. Who knows if the GPUs were boosting as they should. Who knows if the antivirus was running in the background. We can never rule out the possibility that the data is not perfect, even if it comes from Nvidia.



This problem may be the reason why some of the behaviors in the graphs are really weird. Or it's not. I do not know. Maybe the wackiness of the 1440p results come from there. Maybe the weird 4K to 8k speed difference is also caused by that. We’ve talked about how the weird 4K to 8K curve isn't some product of Excel hallucination, but is actually in accordance with what the data shows – and maybe that’s how it behaves in real life. Or maybe there is some slight error in the data, that snowballed into those curves. That’s ultimately something we can’t really know, unless we can have extensive testing from several sources.



The second problem is simply that the Excel predictions take only 4 GPUs into account, limiting the precision. We can already see it in the first picture, the predictions don't completely line up. This is even more apparent if I show you what the predicted curves look like when extending further :





As you see, it goes to shit very, VERY quickly. At around 1.6 it starts becoming unrealistic ; and to give you an idea, on this scale 1=RTX 3080 tensor performance, and the highest actual data from nvidia we have is the 3090 at around 1.2. So it goes to shit real quick. I don’t think that means that what comes before in the curve is useless; the curve still has to intersect at the origin. 0 tensor performance means 0 speed. And there aren’t infinite ways to go to that (I mean yes there are, but shut up you got the point) ; meanwhile, on the right side of the curve Excel doesn’t have any indication of where to go and this explains how the curve can become so unrealistic.



Can this be improved ? Yes.

What we need is more GPUs. Ideally, less powerful and more powerful than what we have now. I have no clue of where to find info about slower GPUs, but I do know where to start with higher end GPUs : I know techpowerup has done some testing on a 4080 for DLSS. This would allow us to see an actual data point at like 1.6, which would go a long way for improving the predictions. Important thing to note is that we can’t rely on Turing to extend the graph, because Turing doesn’t benefit from sparsity, and the improvement gains from that are unpredicatble as we've seen before; on the other hand, Lovelace tensor cores are basically Ampere tensor cores but with FP8. Considering DLSS most likely does not use that, we can use Lovelace cards to extend the graph. Although what would be best would be a lower end card as, you know, we’re trying to predict the performance of Drake. 4080 results may help for that, but 3050 results would be significantly more appreciated, as it would be much more relevant.

In the next version of the calculator, I will try adding 4080 data to make the predictions more accurate, and I will also continue searching for more data that could help us in our quest of predicting Drake DLSS performance.



You have finally reached the end of the technical stuff. Hope you enjoyed it and if you have any question, don't hesistate. There are many things I glossed over, and I’d be happy to give you more details if you want.

 

[Dakhil]

So since ReddDreadtheLead asked, and I'm curious myself, I've created a poll asking when do you think is the launch window for Nintendo's new hardware, which everyone can vote on here.

I also suggest checking out poll #2, which I've made a threadmark for.

 

[Dakhil]

Hi everyone.

After receiving some complaints from FamiBoards moderators about having to continuously increase the character limit for the OP (to 90k characters), I've decided to put all the links for the news and rumours from 2022 and 2021 in Google Docs, which can be accessed here.

I'm sorry for any inconvenience I've caused to the FamiBoard users and the FamiBoard moderators.

 

[Dakhil]

Here we go.

I know this is very late, but here's a long, rough summary from me. Blame the fact I have to work night shifts this week.

Late 2024 Release?

• NateDrake and MVG weren't surprised that Video Games Chronicle has heard about the Nintendo Switch's successor targeting a late 2024 launch since NateDrake has been saying since January 2023 on Nate the Hate that he's heard holiday 2024 as the launch window, which he considered informed speculation on his part due to lack of sources since he only heard from 1-2 sources
• MVG always thought that late 2024 is the launch date for the Nintendo Switch's successor, especially with retail hardware usually being released within 12-18 months of devkits being distributed
• MVG hasn't heard anything from developers about the Nintendo Switch's successor during GDC 2023, which surprised MVG, which made MVG assume Nintendo's waiting to announce the Nintendo Switch's successor or to distribute devkits to key partners later on
• After hearing "reports" of a Spanish dev team receiving devkits, NateDrake reached out to a few of his contacts/sources, and a couple of his contacts/sources (his sources' studios or his sources' colleagues) have confirmed having access to devkits a couple of weeks ago in very recent times, which NateDrake speculates to be as far back as around mid July 2023
• NateDrake thinks retail hardware is going to be released at least 12 months after devkits were distributed, depending on the form of the devkits, whether preliminary or final, with MVG agreeing
• NateDrake thinks there's a possibility the Nintendo Switch's successor could launch in the late September to early October window for 2H 2024, depending on how much stock Nintendo could accrue over the year
• NateDrake also thinks Nintendo could have a monster holiday launch if Nintendo can accrue millions of units in the leadup to Nintendo releasing the launch window of 2H 2024
• MVG thinks the launch of the Nintendo Switch's successor is all about the games ready to be launch with the system (e.g. 2-3 games with Nintendo Switch Online games thrown in, one major first party game with additional games starting to release 3-5 months after)
• MVG also thinks Nintendo won't put a line in the sand and launch the Nintendo Switch's successor in 2024 regardless of if Nintendo has games to launch
• MVG said he wouldn't be surprised if Nintendo pushes the launch to early 2025 depending on where the launch games are at, development wise
• NateDrake and MVG thinks Nintendo pushing the launch to early 2025 would be a disaster scenario, which implies serious problems with software development or hardware production

LCD or OLED Display?

• NateDrake thinks Nintendo needs to release a 3D Super Mario Bros. game at launch, which he thinks would be a premier way to launch new hardware, with MVG agreeing, saying that a 3D Super Mario Bros. game would help make people excited for new hardware
• NateDrake thinks that Nintendo can't afford to struggle with launching the Nintendo Switch's successor, mentioning that Nintendo has historically struggled with releasing the successor to Nintendo's popular consoles (e.g. Wii to Wii U, Nintendo DS to Nintendo 3DS)
• NateDrake thinks Metroid Prime 4 is more like a complimentary game than the reason why people want to buy the Nintendo Switch's successor at launch, as much as NateDrake enjoys the Metroid Prime games, with MVG agreeing, and saying he thinks that Metroid Prime 4's still going to be released on the Nintendo Switch, and won't be a launch title for the Nintendo Switch's successor
• NateDrake thinks that Nintendo releasing Metroid Prime 4 as a launch title for the Nintendo Switch's successor would be a curious decision on Nintendo's part
• NateDrake has heard some whispers (emphasising that this is not confirmation) that the supply line for the Nintendo Switch's successor will begin ramping up production in the early portion of 2024, specifically Q1 2024 (between January and March 2024), which he thinks would further endorse Nintendo targeting a 2H 2024 release, especially with Video Games Chronicle's article mentioning Nintendo wants to have sufficient supply of units for the Nintendo Switch's successor
• Regarding Video Games Chronicle's article mentioning the Nintendo Switch's successor using a LCD display to lower costs, MVG said that people need to remember that Nintendo's starting all over again with the Nintendo Switch's successor as a new generation of hardware, not as a mid-gen refresh ("Switch Pro")
• Therefore, MVG thinks Nintendo will start with the base model all over again as a launch model for the Nintendo Switch's successor with a LCD display, which MVG agrees with
• MVG also thinks Nintendo would probably offer an OLED model for the Nintendo Switch's successor a couple of years later, similar to how Nintendo released an OLED model for the Nintendo Switch a couple of years after the launch of the Nintendo Switch, since that was successful
• MVG thinks Nintendo needs to not increase costs and starting with the base launch model for the Nintendo Switch's successor again, which is going to be re-iterated and improved upon, as a new generation
• NateDrake heard industry whispers, which he considers informed speculation, (and emphasising that this is not confirmation) that the Nintendo Switch's successor's going to launch with a 8" LCD display to reduce costs, as Video Games Chronicle, which he thinks is important to Nintendo since Nintendo's likely to launch the Nintendo Switch's successor at a high price (e.g. $399.99 or $449.99), and recouping costs on manufacturing as soon as possible is important to Nintendo
• NateDrake thinks selling a model with a LCD display would also allow Nintendo to sell a more premium model 3-4 years down the line (e.g. OLED display, etc.), which NateDrake said Nintendo has done with every handheld Nintendo's launched (e.g. Game Boy Advance to Game Boy Advance SP, Nintendo DS to Nintendo DS Lite, Nintendo DS Lite to Nintendo DSi, Nintendo 3DS to New Nintendo 3DS, Nintendo Switch to OLED model)
• NateDrake says that although LCD displays do have crushed blacks and outer, LCD displays, depending on the pixel density, and coming from a 8" display, could be of good quality, and won't look abysmal
• NateDrake mentions that the Nintendo Switch's successor having a 8" LCD display matches with the PlayStation Q having a 8" LCD display
• NateDrake also mentions the Steam Deck, the Asus ROG Ally, etc., also have LCD displays, which are for cost reasons and/or LCD displays look satisfactory enough for manufacturers
• MVG mentions that devkits having a LCD display doesn't necessarily mean that retail hardware also having a LCD display is 100% a lock, especially if developers are using preliminary devkits
• MVG does think Nintendo will launch the Nintendo Switch's successor with a LCD display for cost reduction reasons, and because Nintendo's starting all over again with the Nintendo Switch's successor as a new generation
• Regarding Video Games Chronicle's article mentioning the Nintendo Switch's successor using a LCD display to lower costs, especially since higher fidelity games require increased storage, NateDrake has heard chatter, which he considers informed speculation (and emphasising that this is not confirmation) that Nintendo's planning on equipping the Nintendo Switch's successor with a fairly significant amount of internal flash storage, with the ceiling being 512 GB, which doesn't necessarily mean that the Nintendo Switch's successor won't necessarily be equipped with 512 GB of internal flash storage, but Nintendo does understand that Nintendo needs to offer more than 32 to 64 GB of internal flash storage
• NateDrake thinks that if Nintendo decided to use a LCD display instead of an OLED display, so that Nintendo can prioritise having a fairly significant amount of internal flash storage, he endorses Nintendo's decision, with MVG concurring
• MVG thinks having a larger amount of internal flash storage is a requirement, with 64 GB of internal flash storage not being an option, especially with games having higher resolutions, increased resolution textures, increased asset sizes, increased polygon counts, etc., since everything's going to grow with respect to size, albeit not to PlayStation 5 levels of size
• MVG thinks games being 30 to 60 GB is the standard for games releasing on the Nintendo Switch's successor
• NateDrake says that Nintendo has to recognise that Nintendo needs a significant amount of internal flash storage if Nintendo wants Call of Duty, especially with Call of Duty being 100+ GB in size, with MVG agreeing

New Game Card Tech?

• Regarding the Video Games Chronicle's article mentioning that Game Cards will continue to be used on the Nintendo Switch's successor, NateDrake says that's not a shock to anyone, given the form factor of the Nintendo Switch's successor, and there's no reasonable way to use a disc format, outside of Nintendo's version of MiniDVD for the GameCube, or Universal Media Disc (UMD) for the PlayStation Portable (PSP), which didn't work out for Sony
• NateDrake believes Nintendo will change the tech inside the Game Cards, mentioning that the Nintendo Switch Game Cards currently use 2D NAND memory from Macronix, and there are reports about Macronix developing 3D NAND memory
• NateDrake said he wouldn't be surprised if Nintendo use 3D NAND memory for the Nintendo Switch's successor's Game Cards
• MVG thinks the Game Card format for the Nintendo Switch's successor's Game Cards will change, or has already changed, due to the limitations and issues with the Nintendo Switch Game Cards (e.g. maximum size for the Nintendo Switch Game Cards in concept is 64 GB, etc.)
• MVG also thinks Nintendo revised, enhanced, and iterated the Nintendo Switch's successor's Game Card format to not only allow for larger Game Card sizes that won't be as expensive as a 64 GB Nintendo Switch Game Card, but to increase performance, and to enhance security to make hardware hacking much more difficult, mentioning that the Nintendo Switch was able to be hacked very quickly and very easily very early on in the Nintendo Switch's lifecycle, which allowed Nintendo Switch Game Cards to be easily dumped online, with NateDrake concurring with respect to security

Why Backwards Compatibility Concerns

• Regarding the Video Games Chronicle's article mentioning that some third party developers said that backwards compatibility with Nintendo Switch games could negatively impact sales of the Nintendo Switch's successor's games, NateDrake thinks some third party developers are talking about concerns that consumers won't buy a "Switch 2" game, because consumers are too busy playing older Nintendo Switch games, or potentially the Nintendo Switch version of the same "Switch 2" game
  • NateDrake creates a hypothetical example where Ubisoft releases "Mario + Rabbids Sparks of Hope Deluxe" for the Nintendo Switch's successor, which offers a higher resolution, improved frame rates, and all the DLC included, wanting consumers to buy the game the second time
  • But consumers are not buying, since consumers have already bought all the DLCs for the Nintendo Switch version, and consumers are happy with how the game performs on the Nintendo Switch
• MVG thinks the transition to next generation is Shuntaro Furukawa's biggest test since that has to be done meticulously, or else Nintendo will start to lose customers
• MVG thinks third party developers are trying to say the following based on his reading of what third party developers are saying about backwards compatibility in Video Games Chronicle's article
  • The Nintendo Switch's successor needs to have its unique features and selling points, such as its own games, its exclusive games, its own first party launch games, despite still being part of the hybrid console ecosystem, from a business standpoint of backwards compatibility
  • The Nintendo Switch's successor is not a mid-gen refresh ("Switch Pro")
  • Considering MVG has heard from people about wanting a more powerful Nintendo Switch that plays Nintendo Switch games, the question is are you only asking for a mid-gen refresh at that point, and not asking for anything else, since that implies you want to continue where things are with the current ecosystem, playing your own games
  • If launching new hardware, the new hardware needs its own unique selling points
• NateDrake and MVG believe there are many reasons to have backwards compatibility, such as having goodwill from consumers, and digital backwards compatibility is considered as an industry standard and is an expectation from consumers due to the convenience of going digital, with the understanding and expectation that consumers can easily bring digital games over to new devices
• NateDrake hopes that Nintendo does bring backwards compatibility to the Nintendo Switch's successor, and NateDrake thinks Nintendo not bringing backwards compatibility is a significant blunder on Nintendo's part, which NateDrake will gladly harshly criticise Nintendo for if that's the case
• NateDrake says Nintendo got a pass from the Wii U to the Nintendo Switch since Nintendo changed everything, going from the Wii U to the Nintendo Switch
• And although the Wii U does have backwards compatibility with the Wii, NateDrake mentions that although the process of transferring Wii games to the Wii U wasn't smooth, that was during the advent of digital for all the video game console manufacturers, where video game console manufacturers were thinking of the present and not the future, not understanding the importance of an account system
• NateDrake thinks the Nintendo Switch's successor is a test for Nintendo in terms of if Nintendo views digital as important aspect of the ecosystem, and if Nintendo will do the consumers right by bringing digital backwards compatibility
• Although MVG did mention in the past that there are clear technical reasons why backwards compatibility will be difficult for Nintendo to provide, MVG does believe backwards compatibility is doable
• MVG believes Nintendo's concerned with having people investing in and caring about the new platform, like with the Nintendo Switch
• MVG plays devil's advocate and says the number one reason why backwards compatibility will happen due to the Nintendo Switch's massive install base
• MVG thinks that throwing away the Nintendo Switch product line, server the cord, and drawing a line in the sand, can have negative ramifications with respect to sales for the Nintendo Switch's successor
• MVG doesn't know how backwards compatibility will look like with the Nintendo Switch's successor
• Although MVG doesn't believe backwards compatibility's the number one priority with Nintendo with respecting to launching new hardware, MVG thinks Nintendo's very much aware about people's concerns with backwards compatibility
• And MVG says that speaking about cost cutting, maybe Nintendo could squeeze a Tegra X1 into the new hardware, since Nintendo has done that before with past hardware, with NateDrake laughing
• Assuming that Video Games Chronicle's sources are third party development partners, NateDrake believes development partners should be briefed by Nintendo whether or not the Nintendo Switch's successor has backwards compatibility, so that development partners can plan patches or generally planning upcoming releases to take advantage of the new hardware
• If Nintendo hasn't briefed third party development partners on whether or not the Nintendo Switch's successor has backwards compatibility, NateDrake says Nintendo could still be grappling with how to approach backwards compatibility on the platform
• NateDrake says that devkits technically don't need to be backwards compatibility in terms of functionality, but third party development partners should still be briefed, with MVG agreeing, especially if development partners are using preliminary devkits
• MVG thinks Nintendo's holding back briefing third party development partners about backwards compatibility for later, with details on backwards compatibility being leaked out at some point when there's more smoke about devkits
• MVG thinks people are going to be drip-fed more information about next generation before the announcement of new hardware
• NateDrake and MVG say that there's nothing that can be taken away from Video Games Chronicle's article about backwards compatibility
• If Nintendo didn't brief third party development partners on whether or not the Nintendo Switch's successor has backwards compatibility, NateDrake thinks that's a bad decision on Nintendo's part, since development partners need to know, so development partners can plan patches, and/or including enhancements when played on new hardware
• But NateDrake mentions that depending on the state of the devkits, Nintendo needs to address backwards compatibility within a reasonable amount of time
• If Nintendo updates devkits in firmware in the upcoming months, NateDrake thinks that's something Nintendo addresses, and there's ample time for Nintendo to do right by third party development partners, and have backwards compatibility fully fleshed out when the time comes to launch new hardware

Announcement Timing Thoughts

• MVG said that part of him believes Nintendo could announce new hardware this year, but he isn't sure
• But MVG believes Nintendo needs to get people mentally prepared and hyped about new hardware
• MVG ultimately think Nintendo's probably going to announce new hardware early next year, probably on February 2024
• NateDrake thinks when Nintendo announces new hardware is dependent on launch timing
• If the earliest Nintendo releases new hardware in 2H 2024 is September 2024, then NateDrake thinks Nintendo's going to announce new hardware in April 2024 after the end of the current fiscal year (ending on 31 March 2024)
• NateDrake thinks Nintendo wants to achieve the goal of selling 15 million Nintendo Switch units before announcing new hardware
• Although NateDrake thinks there could be merit to MVG thinking that Nintendo could announce new hardware on February 2024, NateDrake thinks Nintendo wants to wait until the end of the current fiscal year before announcing new hardware
• NateDrake mentions there are many variables that can affect Nintendo's decision of when to announce new hardware, with devkits as one example since Nintendo knows leaks can happen once more devkits are distributed
• The question is does Nintendo want to get ahead of leaks from more devkits being distributed, or wait until when production lines start manufacturing new hardware (e.g. early portion of 2024 or Q1 2024 (January - March 2024) based on whispers NateDrake heard (emphasising that this is not confirmation))
• NateDrake mentions Nintendo informed third party partners on new devkits for the OLED model after the OLED model started mass production
• Therefore, NateDrake thinks Nintendo could announce new hardware between now (since devkits are being distributed) and the end of March 2024
• NateDrake thinks that when Nintendo announces the Nintendo Switch's successor depending on when Nintendo thinks Nintendo have maximum impact in terms of timing
• NateDrake thinks announcing the Nintendo Switch's successor after the end of the current fiscal year is ideal since Nintendo can maximize Nintendo Switch sales and Nintendo won't muddy up any waters, with MVG thinking the logic makes sense
• NateDrake also thinks there's a possibility Nintendo could surprise people by announcing the Nintendo Switch's successor during a September 2023 Nintendo Direct or on October 2023, or when more third party partners have access to devkits
• MVG thinks there's a possibility Nintendo could tease new hardware on September 2023
• NateDrake thinks Nintendo can also wait until third party developers have finished developing games targeting the launch window before announcing new hardware
• NateDrake says expect more information on the Nintendo Switch's successor with Gamescom 2023 and Tokyo Game Show (TGS) 2023 around the corner via unofficial sources (media outlets talking to third party development partners), and maybe official sources via Nintendo, with MVG agreeing
• MVG thinks the remainder of 2023 and early 2024 is going to be exciting in terms of information regarding the Nintendo Switch's successor

 

[Dakhil]

I know somebody already posted the video, but I'm posting the video since I'm doing a long, rough summary of today's Nate the Hate episode.

BotW Switch 2 Tech Demo - Load Time & Resolution

• NateDrake mentioned Gamescom 2023 was the first time since GDC 2023 developers can meet in person, which leads to conversations, which leads to media hearing about information shared during these conventions and industry events, like the tech demos for the Nintendo Switch's successor
• NateDrake also mentioned Gamescom 2023 was a good opportunity for Nintendo and its partners to bring tech demos to demonstrate exactly what the Nintendo Switch's successor could potentially offer when it comes into the market
• And NateDrake mentioned that once Gamescom 2023 was over, two reports were released: one from Eurogamer and one from Video Games Chronicle
• NateDrake started with Eurogamer's report where Eurogamer reported one of the two tech demos Nintendo showed off for the Nintendo Switch's successor was a souped up version of Breath of the Wild, which was designed to the Nintendo Switch's successor's beefier target specs, although there's no suggestion Breath of the Wild's planned to be re-released for the Nintendo Switch's successor
• NateDrake and MVG were confused by people asking if the Breath of the Wild tech demo confirms that the Nintendo Switch's successor has backwards compatibility
• These are the pieces of information about the Breath of the Wild tech demo that NateDrake has heard about, but Eurogamer didn't report on
  • The Breath of the Wild tech demo was running at 4K at 60 fps via DLSS
  • The actual key focus of the Breath of the Wild tech demo was to showcase improved load times
    • NateDrake mentioned that when playing Breath of the Wild on the Nintendo Switch, the time it took from going from the main menu to an in-game save was around 30 seconds
    • Although 30 seconds is not a terrible loading time, 30 seconds feel like an eternity compared to how fast games on the PlayStation 5 and the Xbox Series X|S load from the main menu to an in-game save, which MVG concurs with
  • When running the Breath of the Wild tech demo on the Nintendo Switch's successor (in terms of target specs), what was shown was a smooth transition from the main menu straight to the game, with the load times eliminated
    • MVG was surprised, and asked if NateDrake's talking about a significantly enhanced storage that goes way beyond what's currently possible with the Nintendo Switch, with NateDrake saying confirming yes
    • MVG asked if NateDrake mentioned 3D NAND, with NateDrake confirming that he did talk about 3D NAND a few episodes ago, with "interesting" as MVG's response
• NateDrake re-iterated that the souped up version of Breath of the Wild is only a tech demo
• MVG mentioned that he tweeted on Twitter a couple days ago that when a hardware maker gets developers in a room and shows tech demos, it's really for showcasing specific features of the hardware
• When MVG asked NateDrake if he heard about which version of DLSS was used, considering the conflicting information, with one source saying DLSS 3.5 initially, which was edited out, and another source saying DLSS 3.1, NateDrake heard that DLSS 3.5 was used, but the full feature set of DLSS 3.5 may not be used
• NateDrake mentioned there's a possibility a feature like Frame Generation may not be utilised in the Nintendo Switch's successor, or wasn't utilised in the tech demos
  • NateDrake mentioned there's a little uncertainty on that front for now
• NateDrake mentioned that there's technically nothing preventing DLSS 3.5 from being featured on the Nintendo Switch's successor
• NateDrake hedge his bets that DLSS 3.5 is used for the Nintendo Switch's successor, since DLSS 3.5 is the most advanced form of DLSS, the benefits are definitely there, and DLSS 3.5's something developers can take advantage of
• NateDrake thinks not supporting Frame Generation isn't a big deal since developers are still getting advanced upscaling technology, and developers won't be able to take a game with Frame Generation and radically outperform what developers are getting at the base level
• MVG mentioned the base temporal upsampling is a significant enhancement to what developers currently have on the Nintendo Switch, with the best thing being FSR, which is very simple in terms of upscaling
• MVG thinks the Nintendo Switch's successor's getting DLSS 3.5, a cut down version of DLSS 3.5, or a mobile DLSS 3.5, is pretty big
• MVG says NateDrake's right in saying that DLSS is there for improving performance
• MVG thinks DLSS is one of Nintendo's secret weapons that would give Nintendo a really big advantage going into the next generation of hardware
• NateDrake thinks Nintendo and third party developers can utilise DLSS in significant ways
• And NateDrake thinks DLSS being a base level feature for the hardware will pay dividends for Nintendo and third party developers, since Nintendo can come in with new hardware without having to necessarily compete directly with the Xbox Series X|S and the PlayStation 5 in terms of raw performance to remain competitive, by using DLSS to get a rendered resolution comparable to what the Xbox Series X|S and the PlayStation 5 can output natively without having to render natively, closing the gap in a cheater's type of context, and as a technology of the future
• NateDrake thinks DLSS will make hardware comparisons moot since raw power is no longer a defining factor of hardware
  • NateDrake used the comparisons between the PlayStation 2, the GameCube, and the Xbox as an example, where the PlayStation 2 is advertised as supporting up to 66 million polygons, whereas the GameCube supported 6-12 million polygons, and the Xbox supported 120 million polygons
  • However, the PlayStation 2's advertised 66 million polygons is only theoretical performance, and the PlayStation 2 only supports closer to 5 million polygons in terms of real world performance
  • The Xbox's real world performance never got to be fully utilised due to the PlayStation 2 being the market leader
• The lesson is that raw numbers can't be taken as gospel, and that feature sets, engine support, etc. also has to be taken into account, to determine the true power range of a video game system
• Although the Nintendo Switch's successor won't be as powerful as the Xbox Series S in terms of raw performance, the Nintendo Switch's successor support more modern technologies, which would give the Nintendo Switch's successor a fighting chance to compete with the Xbox Series X|S and the PlayStation 5 in terms of resolution and perhaps frame rates in some scenarios
• MVG thinks the load times on the Nintendo Switch were not the best, but rather okay
• MVG thinks the PlayStation 5 and the Xbox Series X|S spoiled gamers with features like Quick Resume, and most games run very quickly, with some games having an almost seamless gameplay experience, with unnoticeable loading times
• MVG thinks what Nintendo did as a stopgap measure to the noticeable loading times on Nintendo Switch games (e.g. Breath of the Wild, Tears of the Kingdom) is reducing the GPU frequencies and increasing the CPU frequencies during loading, which MVG said he did for some of the games he helped developed, which helped, but is not a viable option
• But MVG thinks Nintendo rightly identified better loading times as one of the priorities for the Nintendo Switch's successor, not to necessarily compete with the Xbox Series X|S and the PlayStation 5, but rather come up with a streamlined approach to loading times, since the reality is there's 4K, better visuals, etc., that are making vgames larger, where 10 GB games are no longer the norm, but rather 40+ GB games are the new norm
• NateDrake thinks that Nintendo showcasing a simple fade from the main menu to the game with the Breath of the Wild tech demo is indicative of Nintendo's and Nvidia's priorities with the Nintendo Switch's successor, where Nintendo and Nvidia are aware that loading times are a relic of the past, and these new standards of gaming need to be adopted in order to compete
• NateDrake also thinks that Nintendo can't come in with new hardware that has loading times where there are two systems with quick loading
• NateDrake thinks that demonstrating Breath of the Wild having that instantaneous fade from the main menu to the game is a strong demonstration of the Nintendo Switch's successor's capabilities, which impressing him more than seeing and hearing that Breath of the Wild runs at 4K at 60 fps, with MVG saying that enhanced visuals are expected
• MVG says that Nintendo focusing on loading performance and the way data is stored is very big and tells him that Nintendo has new tech that people will know about in due course
  • MVG wonders how much the Nintendo Switch's successor would cost, but he decides to continue talking about the reports from Eurogamer and Video Games Chronicle rather than speculate on the price
  • MVG mentioned the time will come for a price prediction for the Nintendo Switch's successor

Matrix Awakens Full RTX on Switch 2 Beyond PS5/Series X?

• NateDrake mentioned the Video Games Chronicle corroborated Eurogamer's report, but went beyond, reporting Nintendo was showcasing The Matrix Awakens Unreal Engine 5 tech demo, which was originally showcased on the Xbox Series X|S and the PlayStation 5 in 2021, running on hardware that has comparable or similar specs with the Nintendo Switch's successor, not on native hardware, citing the Video Games Chronicle's own sources
• NateDrake speculated The Matrix Awakens Unreal Engine 5 tech demo could have been
  • running on a PC with comparable specs
  • a video that Nintendo was using to show what was achieved on devkits
• MVG mentioned that he remembered after downloading The Matrix Awakens Unreal Engine 5 tech demo that he was impressed
• MVG thinks that people making assumptions that the Nintendo Switch's successor is more powerful than the PlayStation 5 and/or the Xbox Series X|S is a very dangerous assumption to make, especially since no one knows which hardware The Matrix Awakens Unreal Engine 5 tech demo was running on, and The Matrix Awakens Unreal Engine 5 tech demo was most certainly running via DLSS and probably via Frame Generation
• MVG re-iterates what NateDrake said about raw performance no longer being the most important factor, saying that being smart in key areas to make the performance run exceptionally well at a handheld resolution or a docked resolution in a hybrid console is the most important, which based on the Video Games Chronicle's report, was done well, with NateDrake concurring
• NateDrake also mentioned the Video Games Chronicle reported on the visuals on The Matrix Awakens Unreal Engine 5 tech demo being comparable to what's shown on the PlayStation 5 and the Xbox Series X|S, with Video Games Chronicles later clarifying that doesn't mean the Nintendo Switch's successor's raw performance is close to comparable with the PlayStation 5 and the Xbox Series X|S
• NateDrake has heard that The Matrix Awakens Unreal Engine 5 tech demo showcased very advanced ray tracing that's identical, if not better, than what's shown on the PlayStation 5 and the Xbox Series X|S in terms of The Matrix Awakens Unreal Engine 5 tech demo
• NateDrake said the Nintendo Switch's successor may have surpassed the PlayStation 5 and the Xbox Series X|S in terms of the implementation of ray tracing in The Matrix Awakens Unreal Engine 5 tech demo
• MVG asked NateDrake to confirm that before the Nintendo Switch was formally unveiled that there was an Unreal Engine 4 demo being showcased, with NateDrake confirming the Elementals Unreal Engine 4 tech demo was showcased
• MVG thinks Epic circling back with Nintendo and showcase new tech on Nintendo's new hardware was always going to be the case, which MVG is very excited to hear, although he doesn't know if the floodgates will open with more Unreal Engine 5 games coming out
• MVG's excited that Epic seems to have been preparing a version of Unreal Engine 5 for the Nintendo Switch's successor, which he thinks developers are excited about as well
• NateDrake said he wouldn't be surprised if the The Matrix Awakens Unreal Engine 5 tech demo was a means of showing that the Nintendo Switch's successor is Unreal Engine 5 compatible, and how much can be achieved when paired with DLSS
• MVG learnt that when Epic is showcasing something, Epic is making a statement to investors, the public, and/or developers, that Unreal Engine 5 is already ready for the Nintendo Switch's successor, and the tools are already be available when developers are ready to make games, unlike other engines (e.g. Unity, etc.)
• NateDrake thinks that showcasing The Matrix Awakens Unreal Engine 5 tech demo demonstrates that the Nintendo Switch's successor's capable at running at a fidelity comparable to the PlayStation 5 and the Xbox Series X|S as reported by the Video Games Chronicle, and how Unreal Engine 5 plays well with Nvidia's technology (e.g. DLSS) that the Nintendo Switch's successor supports
• NateDrake mentions there was uncertainty about the frame rate The Matrix Awakens Unreal Engine 5 tech demo was running at on the Nintendo Switch's successor, with MVG commenting on being surprised if The Matrix Awakens Unreal Engine 5 tech demo was running faster than on the PlayStation 5 and the Xbox Series X|S, which is around 25 fps, not because of performance, but rather to achieve that film aesthetic
• NateDrake mentioned that obviously, developers would naturally ask Epic what's the resolution The Matrix Awakens Unreal Engine 5 tech demo was running at without enabling DLSS, which NateDrake doesn't have answers to, since that's not the point of The Matrix Awakens Unreal Engine 5 tech demo, since the purpose was to show The Matrix Awakens Unreal Engine 5 tech demo with DLSS enabled, and what can be achieved
• NateDrake has heard that developers were impressed by what was shown with The Matrix Awakens Unreal Engine 5 tech demo
• NateDrake thinks that if DLSS is a base feature that every game has access to and can utilise, the Nintendo Switch's successor can be considered a magic box that can punch well above its weight, with MVG concuring
• MVG said as a developer, he's very excited about what was reported on about the Nintendo Switch's successor, and he hopes he can learn more when the time comes

Release & Reveal Timing Discussed at Gamescom?

• NateDrake mentioned there was other talk outside of the two tech demos with respect to the Nintendo Switch's successor that piqued his curiosity, which he couldn't get full clarity
• NateDrake heard a lot of talk at Gamescom of March 2024 with respect to the Nintendo Switch's successor
• MVG was surprised, asking if that's referring to the release date, with NateDrake mentioning there was no clarity if March 2024 is for a release date, a release window, or a reveal date
• MVG admitted March 2024 was not in his bingo card since he thought the Nintendo Switch's successor's coming out in 2H 2024 based on the fact that developers need roughly at least 12 months to familiarise with new hardware
• MVG thinks that talks of March 2024 means that Nintendo needs to announce new hardware really soon, which MVG can't see a reveal happening this year, due to Super Mario Wonder, and the OLED model and Nintendo Switch Lite bundles, releasing next month
• NateDrake said that March 2024 was mentioned during Gamescom more than once, but without specific context
• NateDrake thinks most people would agree that March 2024's referring to the reveal date since that's when the current fiscal year ends where 15 million Nintendo Switch units are projected to be sold, and that gives Nintendo enough time to reveal and plan a 6 month marketing cycle
• If March 2024 is referring to a release, NateDrake mentioned that as MVG said, Nintendo has to reveal the hardware in the near future, or else Nintendo has a very limited marketing window and lead up, which NateDrake said Nintendo feasibly could do, mentioning people will bring up the Nintendo Switch being announced on October 2016, a showcase in January 2017, and release in March 2017
• NateDrake mentioned that people have to remember that Nintendo announced the intention of releasing Nintendo Switch, which at the time was known as the Nintendo NX, in March 2017, on April 2016, almost a full year before launch
• NateDrake mentioned that with any hardware manufacturer typically announces working on or intentions of releasing a successor 10-12 months before release, which Sony did with the PlayStation 5 and was very open with investors, and Microsoft discussed a little bit of the Xbox One's successor before officially revealing the Xbox Series X|S during the Games Awards in 2019, before releasing the Xbox Series X|S 11 months after
• NateDrake thinks that March 2024 being the release date would be a very unprecedented situation where there's no open communication from Nintendo to investors or consumers about having active plans to release a successor in the near future, since right now is the beginning of September, and Nintendo hasn't officially announced any intentions to release a successor to the Nintendo Switch, although that's obvious, given NateDrake and MVG have discussed devkits being distributed, and NateDrake and MVG are currently discussing the tech demos at Gamescom, and Tokyo Game Show is coming around the corner
• NateDrake wonders how Nintendo will play its hand if March 2024 is the release date since no company has played its hand like this before, unless people want to count Sega with its abysmal roll out of the Sega Saturn, with MVG commenting "oof, bad old days"
• MVG said a March 2024 release is hard to wrap his head around, but Nintendo may have its own plans
• MVG asked NateDrake what video game should launch with the Nintendo Switch's successor if March 2024 is the release date, with MVG mentioning the rumoured 3D Super Mario Bros. game, with NateDrake answering maybe Metroid Prime 4 as a cross-gen release, with MVG replying that Metroid Prime 4 is that one video game with no real visibility, with a lot of rumours with unknown verifiability going around about Metroid Prime 4 being a cross-gen game, considering Nintendo has done this before, which is possible
• MVG thinks that dropping a 3D Super Mario Bros. game and Metroid Prime 4, making Nintendo Switch Online (NSO) available to anyone with a Nintendo account and a NSO subscription, and dropping some Nintendo Switch games with patches (e.g. Breath of the Wild) alongside the Nintendo Switch's successor for a March 2024 release would be a very compelling launch
• NateDrake wants to emphasise that the discussion of March 2024 at Gamescom isn't informed speculation, but actually what he heard from conversations about Gamescom, although there's no context about if March 2024 is the release date, a launch window, or a reveal date
• NateDrake thinks March 2024 could be a window where Nintendo asks third party partners to finish up projects (by the end of March 2024) since Nintendo plans on launching hardware a couple months later (e.g. June 2024), which happened to Sony with the PlayStation 5 when Sony asked third party partners to submit games by mid-September 2020 if third party partners are planning a launch day release in November 2020, so Sony can go through log check, the certification process, and there's reasonable time to address bugs and issues
• NateDrake also thinks Nintendo could have planned to release the Nintendo Switch's successor as early as March 2024
• MVG said he's still betting on the Nintendo Switch's successor launching in 2H 2024
• However, he thinks hearing March 2024 being talked about at Gamescom is interesting and definitely throws a wrench
• NateDrake mentioned trying to get clarity about March 2024 being talked about at Gamescom from his sources since Gamescom with no success thus far
• MVG still believes the 2023 is all about the Nintendo Switch with Super Mario Wonder, Super Mario RPG, and the OLED model and Nintendo Switch Lite bundles, with the Nintendo Direct having to be planned for September 2023, with NateDrake mentioning a Nintendo Drake is planned 3 days after this Nate the Hate episode is uploaded, if not announced already by Nintendo
• NateDrake said he would like to agree with MVG, but all the talk about March 2024 at Gamescom gives him great pause
  • The lack of clarity from his sources makes me think a March 2024 release is likely, although he also thinks a March 2024 reveal also makes sense, making this a flip in the coin
• MVG agrees that a March 2024 reveal makes sense
• NateDrake mentions the hardware industry is very fluid and is dependent on the many cogs and gears moving at the right time with respect to mass production, software being ready
  • So because of that, one of the two scenarios: March 2024 being a release day or March 2024 being a reveal date could happen
• NateDrake wonders if March 2024 comes up again at the Tokyo Game Show
  • And if so, he hopes the Tokyo Game Show provides clarity on what Nintendo's intentions are for March 2024 in terms of a release window or a reveal
• MVG mentions Nintendo did have a presence in the business area and asks NateDrake if he's aware if Nintendo will have a similar presence at the Tokyo Game Show, with NateDrake answer he believes Nintendo does have a business meeting room, but not a show floor presence, in the Tokyo Game Show
• MVG thinks there's a lot of smoke, there's only a matter of time people will hear more from Nintendo, and he bets March 2024 is when people will hear more about the Nintendo Switch's successor
• NateDrake mentioned that March 2024 is a point of curiosity that he will be looking at with great focus (pun intended), with MVG being surprised, and NateDrake saying the Nintendo Switch's successor's not called the "Nintendo Focus"

Back-Compat Gamescom Clarity?

• NateDrake has heard a little bit of discussion about backwards compatibility for the Nintendo Switch's successor
• NateDrake hasn't received any clarity with respect to backwards compatibility
• MVG thinks that the Breath of the Wild tech demo reported by Eurogamer confused people to the point people think backwards compatibility is implied
  • MVG think the Breath of the Wild tech demo doesn't really answer any questions about backwards compatibility since the Breath of the Wild tech demo was most likely curated and running native code, not running an emulation layer or wrapper, since the Breath of the Wild tech demo was there to showcase specific technical features to third party developers, so third party developers can understand what the hardware looks like and some things third party developers can leverage when developing next gen games
  • Therefore, MVG thinks nobody's closer to answering the question about backwards compatibility based on the Breath of the Wild tech demo, and he hopes that people will learn more about backwards compatibility at a later date
• NateDrake re-iterated that the Breath of the Wild tech demo is all about showcasing the loading times being erased, going from the main menu to the game, with no splash screen, outside of the high resolution and high frame rate (via DLSS)
• NateDrake mentioned the tech demos at Gamescom were meant to be about showcasing what can be done with the Nintendo Switch's successor and not necessarily about showcasing all the features of the Nintendo Switch's successor
• NateDrake speculates backwards compatibility could be a talking point at the Tokyo Game Show that can reported by the media or that NateDrake hears from his sources
• NateDrake re-iterates that he mentioned in a previous episode (on 2 August 2023) that he and MVG believe the Nintendo Switch's successor needs backwards compatibility and he and MVG would be very concerned if the Nintendo Switch's successor doesn't have backwards compatibility

 

[Anatole]

You're correct in that any attempt to reduce the performance cost of DLSS would impact image quality, but I don't think that's necessarily always a bad thing. When developing DLSS, Nvidia would have had to find a balance between image quality and speed. You can always use a bigger, more complex network (so long as you have sufficient training data) to get better quality*, or a smaller, simpler network to get better performance, and we can assume that DLSS currently represents what Nvidia believes to be the sweet spot, where moving in either direction wouldn't be a worthwhile trade-off.

However, the sweet spot between speed and quality for desktop GPUs isn't necessarily the same as the sweet spot for portable devices with a fraction of the performance. Different trade-offs apply, and what might be considered cheap on a desktop GPU might take an unreasonable portion of the frame time on a low-power console. Even the quality trade-offs may differ, as IQ issues that may be noticeable to someone sitting right in front of a computer monitor may not be as noticeable on a TV screen further away, or a much smaller handheld screen.

I'm sure Nvidia is and will continue to provide the standard versions of DLSS to Switch developers to use in their games, and I don't think there's any free lunch where Nintendo gets a DLSS implementation that's magically faster without any trade-offs, but I do think that there's potentially value, in addition to regular DLSS, to providing a more light-weight version of the model as an option for developers who are comfortable sacrificing a bit of image quality for performance. Whether that's because they're stretching to squeeze in their chosen lighting model and feel it's important enough to sacrifice a bit of IQ by cutting down DLSS time, or because they're targeting 60fps and prefer using DLSS-lite to hit 4K rather than the 1440p output of regular DLSS, or because the limitations of DLSS-lite simply aren't readily apparent in their game (say it has more artifacting around certain high-frequency detail patterns, but they're not present).

* To a certain point. I assume that you'll asymptotically approach "ideal" IQ for the amount of input data you have, and adding excess complexity for this particular task may end up over-fitting or hallucinating, which wouldn't be desirable.

Sorry it's taken me so long to write a reply - work has kept me very busy lately. I know that you understand neural networks, Thraktor, but I've still structured this as a longer explanation post so that people with less experience than you can also engage with my argument. The reason my thinking on this has shifted a bit is that removing layers in a convolutional autoencoder for performance reasons isn't a simple change.

Three features of a convolutional autoencoder​

Working principle of all autoencoders: make the network choose which features matter

An autoencoder is essentially two neural networks concatenated with each other: the encoder, which chooses which features are important, and the decoder, which does reconstruction. By design, the encoder gives up information about the input in each layer; the idea is to force the network to make decisions about which information is relevant to reconstruction.

The more mathematical way to say this is that the encoder learns a representation of the input in a lower dimensional latent space, then the decoder uses the latent space representation to reconstruct the output in a higher dimensional space. This means that, in the encoder, the dimension of each layer needs to be strictly less than or equal to the previous layer; in the decoder, the opposite is true. Another way to think about it is that the encoder is finding an ideal compressed representation of the input, which the decoder is optimized to use for reconstruction.

To translate that into the language of a convolutional autoencoder: the encoder downsamples and learns features about the image, each of which are stored as the channels in a tensor. Those features aren't necessarily interpretable by a human; they have simply been optimized mathematically to pull out the information from the previous layer which would be most useful for reconstruction. Then, the decoder upsamples and uses the learned features to reconstruct the higher dimensional output.

The dimension of the tensor in each layer is the product of the width, height, and number of channels. That dimension, in total, essentially needs to decrease with each downsampling layer in the encoder and increase with each upsampling layer in the encoder. For example, if the width and height both decrease by a factor of 2 while the number of channels is doubled, you've halved the dimensions of the overall tensor.

The computational cost also decreases relative to the "same" network without downsampling and upsampling, but that's just a nice coincidence; the real reason that you need downsampling and upsampling is that you want to force the network to choose which features matter.

Symmetry of an autoencoder: what goes down must come up

Convolutional autoencoders are symmetric overall, and usually only downsample or upsample by a factor of 2 in any given layer. This means that, if you remove a downsampling layer, you also have to remove the corresponding upsampling layer, which means both the feature learning (in the downsampling layers) and reconstruction (in the upsampling layers) are affected.

Receptive field: how to make the network see patterns that span more than 3 pixels

In the convolutional neural network, you can think of the receptive field as being all the pixels in previous layers that contribute to one pixel in a deeper layer. The receptive field is a function of the filter size and stride. For example, what this figure shows is that, with a filter size of 3x3, the receptive field of any point in the second layer would be 3x3 pixels; the receptive field of a point in the output layer would be 5x5 pixels; the fourth layer, if shown, would be 7x7; and so on.

With a convolutional autoencoder, downsampling has a huge effect on your receptive field. Every time the image is downsampled by a factor of 2, you double your receptive field. So if you are iteratively downsampling and filtering say, 4 or 5 times, your receptive field becomes very large.

This is very important for reconstruction because many of the shapes and objects you want the network to learn to reconstruct don't just span one or two pixels. There are some kinds of aliasing, like internal texture detail, that do, but other very common sources, like the edges of geometry, can be very large. What this means for DLSS is that removing a downsampling layer doesn't just affect the complexity of the features the network can learn; it also affects the spatial extent over which the network can learn features and do reconstruction.

Why does this matter for Switch-customized DLSS
​

It matters because it means that the changes you can make to the DLSS architecture have wider ramifications. If you remove one of the downsampling layers, you also have to remove one of the upsampling layers to preserve the symmetry of the network; you also end up reducing the receptive field by a factor of 2, which diminishes DLSS' ability to recognize and reconstruct longer range patterns; and you remove one of the steps in which the network is forced, through downsampling, to make mathematical decisions about which information it needs to preserve for reconstruction.

This means that removing a layer is not just a simple tradeoff, where you can tweak the dimensional knobs of the network a bit to reduce the computational cost. It has major ramifications for the spatial scale of the information you can reconstruct.

So what changes can we make?
​

There are still some options that can make the network a bit lighter. Maybe the easiest way to explain this is to show a few different examples of convolutional autoencoder architectures.

Facebook neural supersampling

This is the architecture from the Facebook neural supersampling paper that we've all discussed quite a bit. The core reconstruction network is a convolutional autoencoder, where you can see some of the features I left out of my "simple convolutional autoencoder" explainer post. In particular:

• Instead of using stride size and convolution/transpose convolution for downsampling and upsampling, they use pooling and upsampling (from other experience, I believe it's probably max pooling and bilinear upsampling, respectively, but it's not actually mentioned in the paper).
• There are two layers between each downsampling/upsampling step. However, these layers don't violate the "rules" of autoencoders; the encoder never increases in overall dimension, and the decoder never increases in overall dimension. However, having multiple layers does let you somewhat increase the complexity of the features you can learn; in multilayer perceptrons, I once saw a proof that adding an additional layer has a combinatorial effect on the complexity of the functions you can learn. That's the reason that deep learning, with many layers, is more effective than a shallow neural network with one enormous layer, even though the shallow network of an arbitrarily large size can theoretically learn an arbitrarily complex function. But I'm digressing.
• That means you could strip out those "extra" layers between downsampling and upsampling steps if you wanted to reduce computational cost. But like I said, each layer lets the network learn significantly more complex features. It can have a big impact on image quality.
• The Facebook paper also includes an additional network which they refer to as Ours-Fast, which reduces the number of output channels (e.g., the number of learned features) of each layer by 50%, which decreases the total run time from 24.4 ms to 18.3 ms (on a Titan V; this paper is getting old). The image quality does decrease, but is still quite close in structural similarity and PSNR to the full size network (it's Table 3 in the paper, if anyone is checking). This is perhaps the primary optimization that a custom version of DLSS could make.
• The network has skip connections between the input layers and the output layers of the corresponding size. This is because downsampling always removes high frequency information (in the spatial frequency sense). Using skip connections allows the network to only learn the difference (in statistical terms, the residual) between the input and output, preserving high frequency information.

Deep DIC

Digital image correlation is to solid mechanics what optical flow is motion in image processing; it tracks the displacement of each particle in a random pattern (think of spray paint) deposited on the surface of a sample and uses them to generate displacement and strain fields. One of the problems in the field is that discontinuities, like cracks in fracture, don't behave very well with DIC. The idea between this Deep DIC paper, which is a collaboration between Northwestern and Ford research, is that you can train a neural network to learn how to reconstruct displacement and strain fields from synthetic (computer generated) DIC data which contains discontinuities.

They train two networks, one for displacement and one for strain. The input to these networks is two images: a reference image and a deformed image. The output for the displacement network has two channels (x and y displacement) and the strain network has three channels (two normal strain components and a shear strain component).

Here are some of the features of this network:

• One layer uses max pooling, but all the others use convolution and deconvolution. The max pooling layer occurs early on to force the network to make a decision about what information to propagate forward.
• Again, we see the symmetry of the encoder and decoder, with skip connections to propagate higher frequency information through the network. This is important because one of the main goals is reconstructing discontinuities, which are always high frequency.
• The network follows the autoencoder rules about dimensions of the encoder always decreasing and the encoder always increasing, except for the last layer of the encoder in the displacement network. It also follows the common convention of always doubling the number of channels and halving the height and width in the encoder, and vice versa in the decoder.

Intel supersampling and denoising

This network is equivalent to DLSS ray reconstruction, doing supersampling and denoising with a single network. I want to make a separate post about this at some point, but I haven't had the time. In particular, it does some really interesting things with its loss function which may explain some aspects of DLSS. But I'll save that for later.

This network is actually several blocks chained together, so it's much more complex. But to summarize the features:

• The input block does an initial projection step (a normal layer without an activation function), then learns some initial features and average pools the output to prepare the input for the feature extractor. There's a skip connection from the input block to the output block to pass through the high frequency information.
• The feature extractor is a version of a convolutional autoencoder, specifically a modified U-Net (which is a standard autoencoder-type architecture in image segmentation). The encoder is shown in blue and the decoder in green. The classic U-Net doubles the number channels with every step in the encoder and halves the number of channels in the decoder; the U-Net has four downsampling steps, so the innermost layer in the encoder ends up with 1024 channels (64 * 2^4).
• From the perspective of DLSS modifications, the most important thing to see is that this feature extractor network no longer simply doubles the number of channels with each downsampling step. The Intel network still downsamples four times, but the innermost layer only has 256 channels, significantly decreasing the computational cost while still maintaining the receptive field and forcing the network, through downsampling in the encoder, to make choices about which information to send through to the decoder for reconstruction. The "basic rule" of an autoencoder, that the dimension of each layer decreases in the encoder and increases in the decoder, is still followed.
• This is what I mean when I argue that it's unlikely that any modified version of DLSS on the Switch 2 would remove downsampling/upsampling steps. Even in a modified network, these steps are a key feature of the autoencoder. Removing them would have serious ramifications for the output image quality.
• Again, skip connections are used to pass through high frequency information from the encoder to the decoder. This network uses max pooling for downsampling in the encoder and bilinear upsampling in the decoder.

That's a lot of bullet points about the autoencoder portion of the reconstruction. But maybe the most interesting/unusual part of Intel's approach here is the filter paths, which handle reconstruction of specular/diffuse lighting and albedo/normal. This looks like a convolutional network in architecture... but it's not. It might not even qualify as a neural network!

• In a convolutional network, filters of a certain size are run over the input from the previous layer. Often, these filters are 3x3, but not always. These same parameters are shared all over the image to pull out relevant features. The weights of these filters are trained during the mathematical optimization of the network.
• The filters in the "filter paths" section of this network are not convolutional. They aren't shared all over the image. They aren't even learned, in the normal sense of machine learning. Here's a verbatim quote from the paper to drive the point home: "there are no learnable parameters in the filter paths."
• Instead, they're per pixel filters over the entire image space which are predicted by the kernel predictor (kernel is another word for filter) as the feature extractor runs. What this means is that, as each layer of the decoder runs, the input for that layer is also filtered by the kernel predictor to get per pixel filter values for every point in the diffuse/specular maps. To me, that's wild! I hadn't seen anything like this before, although they cite papers that have done similar work, so I guess there's a precedent.
• There's still skip connections, and the output of each layer in the network still serves as the input for the next layer. And there's a a ReLU activation function too. So in every sense, the diffuse/specular filters are like a convolutional autoencoder, except that the weights aren't trained or shared across the image.
• The downside of this approach is that, since this isn't convolution, it doesn't map to matrix multiplication in the same way that convolution does. For that reason, Intel runs this part of the network using shaders in Vulcan rather than using TensorRT. Calls to execute the actual neural neural network layers, on the other hand, are handled with TensorRT. Table 4 summarizes how much time is spent on each execution path (4.08 ms in TensorRT and 1.11 ms in Vulcan).
• Albedo/normal are not treated with an autoencoder-like architecture like diffuse/specular are. Instead, per-pixel kernel weights are predicted in the output block. The reason that you would want to use the autoencoder-like architecture is to increase the receptive field and the complexity of the features that you learn; however, the paper claims that you don't need this for albedo/normal. I'm not sure if I'm representing the subtleties of this point well, so here's the direct quote: "We do not predict kernels for auxiliary buffers(albedo and normal) at this stage [during feature extraction], since a large receptive field is not necessary for aliased signals."
• There are some other interesting aspects here about temporal filtering that I need to understand better to write about; I may try to do this another time, when I've spent more time with it. My current understanding is that the current frame and the previous frame, warped with motion vectors, are handled with different filters in some (maybe all?) cases.
• Finally, the filtered specular, diffuse, and albedo are combined into a composite and bilinearly upsampled to the output resolution, then concatenated with the features from the output block, and filtered with another with another per-pixel filter at the output resolution, which generates the output image.

That's a lot more bullet points than I meant to write, but there are so many interesting aspects of this network. I'd really recommend reading the paper to anyone who's interested in how DLSS ray reconstruction works.

tl;dr
​

Yeah, me too. This post got way off track. Sorry about that; hope it was interesting! I fucking love this stuff.

The takeaway point, as far as Switch 2 is concerned, is this: convolutional autoencoders are tricky. You can't strip out the downsampling or upsampling layers, because it takes away the main power of a network like this: learning what features matter over a large receptive field.

There are some changes that Nvidia could still make to DLSS for Nintendo, like decreasing the number of channels per layer. But I'm increasingly of the opinion that they won't do this. Whatever the specifics of the DLSS architecture look like, they've been carefully selected by Nvidia to balance all of these considerations. Removing layers or decreasing the number of channels always has downstream effects in a neural network, but in a convolutional autoencoder, they're particularly prevalent.

Honestly, I'm now falling more into the camp that I've seen Oldpuck express before; if there's an issue hitting 4K with DLSS, just have DLSS output to 1440p and then use another upsampling method to get the rest of the way. Maybe I'll be wrong about this! I am 100% sure that the kind of magic, "hardware-optimized" model that some posters seem to expect does not exist, but there are some small tweaks you can make to the architecture to get, perhaps, up to 30-40% more performance. Either way, I hope this post can lead to a conversation that's grounded in the specifics of how a neural network model for this kind of thing works.

We out.

Citations
​

Facebook neural supersampling: https://research.facebook.com/publications/neural-supersampling-for-real-time-rendering/

Deep DIC: https://www.sciencedirect.com/science/article/pii/S0924013621004349

Intel supersampling denoising: https://www.intel.com/content/www/u...rally-stable-denoising-and-supersampling.html

My previous post about some of the basics of a convolutional autoencoder: https://famiboards.com/threads/futu...-staff-posts-before-commenting.55/post-450982

My post on aliasing vs noise: https://famiboards.com/threads/futu...-staff-posts-before-commenting.55/post-783007

 

[LiC]

Since I had it in front of me, here are the differences in hardware definitions between T234 and T239 (if we assume certain files were up to date), based on LIC definitions (no relation; stands for Legacy Interrupt Controller). This is a pretty dry list and more for my own reference, so peruse only with discretion.

The only unsolved thing here is the PPC and PPC2DISPLAY defines for T239. Anyone have any guesses what those are?

Edit: PPC is not a very unique acronym, but assuming what I'm looking at is the same usage, then PPC encompasses the PES (Shared Primitive Engine), so maybe it stands for Primitive Processing Cluster? Nvidia doesn't use that term publicly, but it could refer to that part of the pipeline. Although if that's the case, I still don't know what PPC2DISPLAY means, since the primitive pipeline is not sending anything to the display.

 

[Dakhil]

New Nate’s video

Tokyo Game Show Rumors -- Xbox, Panic Button & More

• NateDrake mentioned that Nash Weddle posted a miniature Twitter thread that made mentions of some discussion points of the Nintendo Switch's successor at the Tokyo Game Show 2023, including Metroid Prime 4 being shown to developers outside of Retro Studios, and Xbox seeing the Nintendo Switch's successor at Tokyo Game Show 2023
  • NateDrake found Nash Weddle's Metroid Prime 4 claim bewildering since various departments at Nintendo Co. Ltd. (Japan), Nintendo of America, Nintendo of Europe, etc., are obviously getting fairly consistent updates on Metroid Prime 4's status to prepare marketing plans
  • Hence, NateDrake thinks Nash Weddle's Metroid Prime 4 claims are safe guesstimates
  • And although Nash Weddle's Xbox claim is safe in many ways since Microsoft is a development partner for Nintendo and Sony, especially with Microsoft successfully acquiring Activision Blizzard, and Microsoft being the publisher for Minecraft, Ori and the Blind Forest, Banjo-Kazooie, etc., Tokyo Game Show 2023 is probably not the right venue to have hardware demonstrations, especially since Microsoft's and Nintendo of America's offices are so close, and Microsoft can literally walk to Nintendo of America's office
  • NateDrake said Gamescom, GDC, and E3 to a certain extent, could be possible venues for hardware demonstrations, but not the Tokyo Game Show
• NateDrake thinks that Nash Weddle is hedging bets that no one can outright confirm and outright deny Nash Weedle's claims
• NateDrake and MVG think Nash Weedle's claims are far fetched
• MVG has heard chatter about there being a technical demonstration for the Nintendo Switch's successor at Gamescom 2023, which he called Gamescom 2023's worst kept secret
  • But MVG heard absolutely nothing about the Nintendo Switch's successor at Tokyo Game Show 2023
  • And since there were no reports from major outlets, MVG doubted Nash Weddle's claim that Xbox was shown the Nintendo Switch's successor during Tokyo Game Show 2023
• NateDrake thinks Microsoft is more likely to meet up with Nintendo at Tokyo Game Show 2023 to discuss upcoming deals regarding Minecraft, Nintendo Switch Online, etc.
• MVG has heard that nothing really happened at Tokyo Game Show 2023
  • MVG thinks that if something did happen during Tokyo Game Show 2023, it's very airtight, and only a few handful of people probably know about it
• When NateDrake asked his sources attending Tokyo Game Show 2023 if the Nintendo Switch's successor was a conversation point, his sources said there were no whispers or chatter on the showfloor about the Nintendo Switch's successor
• NateDrake thinks this shouldn't necessarily be a concern since if Nintendo discussed the Nintendo Switch's successor at Gamescom 2023, Nintendo might not feel the need to repeat it at Tokyo Game Show 2023
  • NateDrake thinks there could have been off-site meetings with partners and other individuals about the Nintendo Switch's successor, which weren't discuss outside of those meetings
  • The information from those meetings could have been insignificant, or be a repeat of what was shared at Gamescom 2023
• NateDrake said that if there were claims, and there's no corroboration from others, then the claims don't really have a foundation to stand on
• NateDrake mentioned Nash Weedle also claimed that Panic Button was a developer of conversation at Tokyo Game Show 2023 for the Nintendo Switch's successor
  • Although Panic Button is a talented outsourcing studio who did a fantastic job with the commissioned Nintendo Switch ports projects, to NateDrake's recollection, Panic Button hasn't been commissioned by any Japanese developers to port games to the Nintendo Switch, but rather Western developers (e.g. Bethesda)
• MVG thought Nash Weedle's claim didn't make any sense since as NateDrake mentioned, Panic Button worked with Western developers, not Japanese developers
  • MVG could see Panic Button potentially being at the technical demonstrations at Gamescom 2023
  • MVG mentioned that although Panic Button for the most part does quality work on the Nintendo Switch, with the exception of Apex Legends, which is mostly likely due to Panic Button being on a serious crunch schedule, that doesn't necessarily mean Panic Button is working on games for the Nintendo Switch's successor right now
  • MVG also mentioned that there are also other port studios that do quality work on the Nintendo Switch, but are people going to assume the other port studios have devkits for the Nintendo Switch's successor?
• MVG mentioned that Panic Button is a small port studio that focuses on Western games, which is why MVG doesn't believe Nash Weedle's claim that Panic Button is a developer of conversation at Tokyo Game Show 2023
• NateDrake said that he doesn't think Panic Button is a developer of conversation at Tokyo Game Show 2023
• But NateDrake and MVG do believe Panic Button will eventually get devkits for the Nintendo Switch's successor, so Panic Button can assist other studios bring ambitious games developed for PlayStation 5 and Xbox Series X|S to the Nintendo Switch's successor
• NateDrake thinks that Nash Weedle used logic and reason, and then attach a well known studio that's done well on the Nintendo Switch, to make the claims seem credible, although the claims are in reality simply whispers in the wind
• NateDrake mentioned that Nash Weedle also claimed that the Yazuka games are coming to the Nintendo Switch's successor at launch, also mentioning that Sega and Yazuka producers talked about one of the reasons why the Yazuka games didn't come to the Nintendo Switch due to the Yazuka games weren't a good fit for Nintendo's platform
  • NateDrake wonders if Nash Weedle is once again hedging a bet that Sega has seen that the Nintendo Switch could be a viable platform for the Yazuka franchise
• NateDrake did a surface level investigation, and found from a good source, although not with 100% certainty, that the Yazuka games aren't planned to be released on the Nintendo Switch's successor presently
  • NateDrake said that the decision isn't finalised, or Sega's having internal discussions, or Sega's still evaluating the plans
• Nash Weedle's claims of the Yazuka games coming to the Nintendo Switch's successor gave MVG pause since porting the Dragon Engine via the Nintendo SDK sounds like a massive undertaking on a technical level, although MVG said there's a possibility Sega's planning on porting the Dragon Engine in the future, and the Yazuka games could do very well overall on the Nintendo Switch
  • MVG mentioned that Sega was criticised for Like a Dragon running on 720p on the Xbox Series S and 1080p on the Xbox Series X, which MVG attributes to the Dragon Engine
  • MVG thinks bringing the Dragon Engine to the Nintendo Switch, and then optimising for Arm, involves a lot of work
• MVG does mention there's a possibility Sega's looking into or currently working on bringing the Dragon Engine to the Nintendo Switch due to how successful the Nintendo Switch is, sales wise
  • But MVG thinks that there are technical issues that needs to be addressed before bringing the Yazuka games to the Nintendo Switch, which is a resource and money investment (e.g. $10 million - $100 million cost)
  • And MVG thinks that if Sega wanted to bring the Yazuka games to the Nintendo Switch, Sega would have started development a few years ago
• Although MVG doesn't believe the Yazuka games are coming to the Nintendo Switch's successor at the moment, he thinks the Yazuka games could come in the future
• NateDrake mentions that Sega likes to go through Sega's backlog of games, but whether or not a Yazuka game is coming, depends on if Sega believes there's a market and an audience for the platform
• NateDrake said he wouldn't invest much attention or money on Nash Weedle's claim of the Yakuza games coming to the Nintendo Switch's successor at launch
• NateDrake thinks Nash Weedle's claims are too vague and safe to have any solid backing

Release Timing - September or November

• NateDrake mentioned there were rumours about there existing both a physical media and digital SKU, and a digital only SKU, and a launch happening at the end of September or the first week of November (of 2024)
  • NateDrake also mentioned many people dismissed the rumours since the rumours mentioned NG as the codename
• NateDrake mentioned receiving information a couple of weeks prior that largely matched the rumours, which he didn't have time to corroborate with his sources
  • But NateDrake found some of the information to be dubious since he didn't think development partners would be told a year in advance the specific day of launch, which the rumours claimed was the 24th September 2024, or the 5th or 6th of November 2024, since the specific launch day is usually ironed out 6 months from launch, not 12-13 months from launch
  • NateDrake mentioned Nintendo are probably writing down specific dates internally, which are erased and changed periodically
• MVG also believes that the rumours are far fetched and aren't true
  • MVG thinks people are taking the rumours at face value since nobody's really challenging the validity of the rumours, which shouldn't be the responsibility of the viewers/listeners
  • And MVG doesn't know where the specific dates come from, since there's no way a specific launch day is ironed out that far in advance
• Therefore, it's hard for MVG to take the rumours seriously at face value
• NateDrake thinks the rumours are too specific, date wise
• NateDrake thinks claiming a September 2024 launch in September 2023 is a bold claim since that's making an assumption that there's no issues whatsoever on the supply line, etc.
  • NateDrake mentioned that for example, if someone claimed on April 2024 that the console is releasing on the 15th of October 2024, that's believable, since 6 months is typically a lead time for console manufacturers
• NateDrake thinks the person who's reporting the rumours could be overzealous
• And MVG think that people should use common sense when it comes to rumours

Two Sku - Digital & Physical Model?

• NateDrake mentioned that the rumours specify that a physical media and digital model would cost $450, and a digital model would cost $400, which NateDrake thinks warrants more attention
• NateDrake said if he's Nintendo, he would definitely entertain the idea of having a digital only SKU at a more affordable price alongside a physical media and digital SKU, since that's what Sony and Microsoft are doing, digital being the trend in the video game industry, and Nintendo's pushing towards digital purchases with the Nintendo Switch Game Vouchers, with a digital only SKU only amplifying that
• MVG agrees with NateDrake and thinks Nintendo will eventually phase in the digital only SKU as the main SKU, although MVG and NateDrake hate that
  • Although it pains MVG to say this, he thinks Nintendo has to cater to digital only customers, because if they are provided a Game Card slot that they never use, then it's a waste of money
  • So making a digital only SKU appealing in terms of price and features makes a lot of sense
• Although MVG doesn't know if Nintendo will launch with two SKUs, he could see Nintendo doing that
• When NateDrake heard the two SKUs portion of the rumour, he said he couldn't dismiss that portion since he could see Nintendo releasing two SKUs since Nintendo can control the ecosystem more strongly, and make more money from Nintendo Switch Online
  • Since Nintendo has been burdened with piracy due to the launch Nintendo Switch being easily exploitable, which Nvidia's trying to prevent with the Nintendo Switch's successor, NateDrake thinks a digital only SKU could enable Nintendo to prevent street date violations
• Although NateDrake doesn't think Nintendo will go all digital with the Nintendo Switch's successor, he thinks Nintendo could go all digital 1-2 generations after the Nintendo Switch's successor, especially with 60% of sales in Japan are physical
• NateDrake doesn't think a $50 difference is not enough to incentivise customers to buy a digital only model, but rather at least a $100 difference could be
• NateDrake doesn't know if the two SKUs rumour came from Nintendo's development partners, or it's a safe assumption based on the trends in the video game industry
  • The person who sent the information to NateDrake was unknown to him, and he wondered if the person who sent him the information is the same source for the rumours
• MVG thinks the two SKUs rumour has the most credibility thus far

New Backward Compatibility Rumor

• NateDrake mentioned a week ago there was a rumour that suggested a digital only SKU will not have digital backwards compatibility, but a physical media SKU will have backwards compatibility via physical medial, which baffled him
  • If he was Nintendo, he would enable digital backwards compatibility, but disable physical media backwards compatibility, for a digital only model
• MVG said he really struggled to try to make sense of this rumour, especially the digital only SKU not having digital backwards compatibility
• MVG wonder what happens if he wants to purchase games from the Nintendo eShop in his digital only SKU
• MVG couldn't make any sense of there being a feature/flag baked into the firmware that prevents digital only SKU customers from playing old games from the Nintendo eShop due to no Game Card slot, since in that case, the two models are separated by a major feature, which is taken away from the more affordable SKU, but is present in the more expensive SKU
• MVG thinks backwards compatibility is an all or nothing situation since there's one set of firmware across all the products
• Although MVG thinks NateDrake's hypothetical scenario of Nintendo disabling physical media backwards compatibility and retaining digital backwards compatibility makes more sense, he still couldn't make sense of it, since he thinks Nintendo will offer 100% backwards compatibility or not, which NateDrake agrees with
• If Nintendo doesn't offer backwards compatibility, MVG thinks Nintendo would offer patches/upgrade paths for playing old games anyway
• As NateDrake was reading the transcript of the video that reported on the rumour of the digital only SKU having no digital backwards compatibility whereas a physical only SKU has backwards compatibility, he was even more confused
• NateDrake thinks if there's any validity to the claim of the digital only SKU having no digital backwards compatibility, but the physical media does, then that's the worst implementation of backwards compatibility from any company, which he can't see Nintendo doing
• (To (Shpeshal) Nick) MVG said sorry, but the rumours aren't right, since Nintendo's either all in on backwards compatibility, or not, as shown with the Xbox Series X|S, where backwards compatibility is across the board

Recent Patents - Mean Anything?

• NateDrake mentioned there was a patent that showed a hardware layout that people think Nintendo will adopt for the Nintendo Switch's successor, which has the usual ABXY buttons, a single analog stick, and has a form factor similar to the Nintendo Switch Lite
  • NateDrake doesn't think the patent suggests what the Nintendo Switch's successor's form factor looks like, especially with a single analog stick being a nightmare for backwards compatibility
• NateDrake reiterated that if Nintendo released a patent publicly, it's a concept that Nintendo is/was exploring that Nintendo's not likely to use in the future since Nintendo likes to keep secrets for as long as possible
• NateDrake doesn't know why people keep bringing up public patents from Nintendo as an indication of what Nintendo plans for the future, bringing an oval shaped system patent from 2016 as an example, since the Nintendo Switch never adopted that form factor
  • NateDrake pleads people to stop reporting on patents
• MVG mentioned never looking at patents
  • Although patents are interesting to chat about, there's really nothing from patents, and a lot of people don't know what they're looking at with respect to patents, while trying to make sense of when patents were renewed and submitted
• MVG thinks trying to fit patents into something is a very dangerous game since that makes people believe that Nintendo's only working on the Nintendo Switch's successor, when in reality, Nintendo gets involved in all sorts of things
• NateDrake does think some patents are an interesting conversation point of what if Nintendo did utilise this gameplay concept, or implement this oval form factor
• NateDrake mentioned one of the Wii controller prototypes (in a patent) was just a star, and that Nintendo at one point patented a horse saddle, which he's glad didn't came to market
  • MVG joking asked if that horse saddle was for MC Horse, and Nate answered no, with MVG laughing

Custom SoC & Feature Set

• NateDrake mentioned he was chasing information for a week that he was given permission to share
• NateDrake has heard that the Nintendo Switch's successor will have ray reconstruction
• MVG found that interesting, which made him think that's one of the reasons why third party developers said that the ray tracing in The Matrix Awakening demo looks good and impressive, with NateDrake mentioning ray reconstruction being a contributing factor
  • NateDrake iterated that the results and the quality of the ray reconstruction will depend on the denoiser, and won't be universally applied, but rather by a game by game basis
• NateDrake mentioned that with the Nintendo Switch's successor's SoC supporting ray reconstruction, the Nintendo Switch's successor will be able to outpace the PlayStation 5 and the Xbox Series X|S in terms of ray tracing easily, especially as Nvidia further develops the (ray reconstruction) technology, and it can be updated with firmware, which guarantees ray reconstruction can get better on the Nintendo Switch's successor, and that the Nintendo Switch's successor have state of the art technology in comparison to the PlayStation 5 and the Xbox Series X|S
• When MVG asked for clarification about outperforming the PlayStation 5 and the Xbox Series X|S, NateDrake answered that's strictly for ray tracing and no other category, with MVG mentioning he's seen lots of different takes of how powerful the Nintendo Switch's successor is
• MVG thinks the Nintendo Switch's successor outpacing the PlayStation 5 and the Xbox Series X|S in terms of ray tracing makes sense since Nintendo's going into a new generation, Nvidia's Nintendo's partner, and Nvidia does have very good state of the art ray tracing techniques
  • MVG mentioned the Nintendo Switch’s successor's hardware is more modern in terms of when the SoC’s been developed and ready
  • Therefore, getting a new generation of ray tracing on consoles makes sense to MVG
• MVG mentioned talking about the PlayStation 5 Pro having more advanced ray tracing as the potential reason for the PlayStation 5 Pro existing
• But having new hardware from Nintendo that has advanced ray tracing does make sense to MVG since this is 2024, not sometime in the past, and he expects more advanced ray tracing techniques with the Nintendo Switch's successor
• NateDrake mentioned the Nintendo Switch's successor's SoC is a custom chip with a custom feature set
  • Therefore, there's no chip(s) that can be used for a 1:1 comparison
• NateDrake mentioned that back in August 2023, although he heard the version of DLSS the Nintendo Switch's successor uses is DLSS 3.5, it may not be feature complete with DLSS 3.5 on PC, with one feature he heard was being omitted is Frame Generation
  • But since the Nintendo Switch's successor's using a custom feature set, it can support many current technologies Nvidia's bringing to PC, such as ray reconstruction, which is a fairly new release
• NateDrake said that shows what the Nintendo Switch's successor can have in terms of feature sets, which is basically anything Nvidia wants to put on it at the moment, which allows Nvidia to build a system that performs very admirably for the foreseeable future
• NateDrake mentioned the SoC is quite well built for Nintendo by Nvidia that has ray reconstruction and other feature sets, which ensures the Nintendo Switch's successor won't be outdated when 2026 comes, and stands the duration of the new generation
  • Therefore, NateDrake thinks Nintendo's very fortunate to have Nvidia as a partner who continues to make huge strides in the mobile market with its technology, and pushes the envelope in terms of PC and mobile technology
• Although NateDrake mentioned that AMD's a competent company in its own right, crafting fine hardware for the PlayStation 5 and the Xbox Series X|S, Nvidia's far ahead of AMD in terms of AI technologies (e.g. DLSS, ray reconstruction) and many other things
• NateDrake said raw power doesn't really matter anymore, but rather the feature sets supported, and how much RAM the Nintendo Switch's successor has, are important, since ray reconstruction and DLSS 3.5 allow for high image quality at only a fraction of the raw power compared to other platforms
• MVG agrees, saying that bar charts in terms of the largest TFLOPS, the largest performance, etc., are no longer important
  • MVG said that key updates in key areas (e.g. loading speeds, enhanced visuals that look and run faster, faster SoC with much improved memory bandwidth and increased memory (capacity)) allow for a system that's not only more powerful than the previous generation, but also can still run with a battery life of 2-6 hours
  • MVG mentioned that Nintendo's very careful in providing a system that has some semblance of decent battery life
• Therefore, MVG thinks Nintendo kind of has the golden ticket with Nvidia as Nintendo's partner, with Sony and Microsoft, Nintendo's competition, are partnering with AMD
• MVG mentioned that when Nintendo partnered with Nvidia on the Nintendo Switch, people were hesitant about how well Nintendo's and Nvidia's partnership would work
  • But Nintendo's and Nvidia's partnership worked out really well for Nintendo, and Nvidia hasn't stop iterating on mobile SoCs
  • Therefore, MVG thinks what people will ultimately see from the Nintendo Switch's successor will be very impressive
• MVG also thinks the Nintendo Switch's successor will be a fast system, but only due to the new feature sets and technologies being introduced, since it won't be competing in terms of specs in any other way, except for being a very optimised and fast hybrid system that makes games run faster, look better, and load faster, which are the key areas MVG believes Nintendo needs to focus on, which is obviously the case
• NateDrake thinks this is a case of working smarter, not harder, which fits Nintendo's philosophy exceptionally well, since Nintendo's not trying to compete in terms of brute force and raw computation, but rather making its platform efficient, which is a smart direction
  • NateDrake thinks the (video game) industry will be heading in that direction in the upcoming generation, thanks to Nvidia changing the game, showing technologies, such as DLSS, and other companies are developing competitors to DLSS (e.g. AMD with FSR), where brute force won't be carrying any type of computation, but rather AI will be rendering, which allows for outperforming what the raw specs say
• NateDrake's interested to see how the Nintendo Switch's successor's SoC performs
• NateDrake thinks it would be an interesting day if Nintendo's the market leader in terms of ray tracing with the Nintendo Switch's successor, Sony's the market leader in terms of resolution with the PlayStation 5 Pro, and Microsoft's the market leader in terms of frame rates
• MVG mentioned being excited for the future, thinking that Nintendo will have a very successful decade, and maybe beyond that
• NateDrake's looking forward to seeing the first trailers of games coming to the Nintendo Switch's successor when Nintendo announces it, considering that the mobile market has advanced with Resident Evil Village coming to iPhones out of all platforms

 

[Dakhil]

Anyway, before anyone thinks T239's automatically outdated, I highly suggest reading @necrolipe's post below.

"If the Switch 2 launches in 2025, its SoC will be outdated"

Here I leave an example of research and development of the next Xbox system

Look at the time it takes between the decision and design of the SoC and the launch of the final product