D
Deleted member 1324
Guest
TAA is the requirement.
We will see about the GPU power. 2.3tf is still above what I personally expect.
-
Hey everyone, staff have documented a list of banned content and subject matter that we feel are not consistent with site values, and don't make sense to host discussion of on Famiboards. This list (and the relevant reasoning per item) is viewable here.
StarTopic Future Nintendo Hardware & Technology Speculation & Discussion |ST| (Read the staff posts before commenting!)
- Thread starter Dakhil
- Start date
We will see about the GPU power. 2.3tf is still above what I personally expect.
- Pronouns
- He/Him
Actually I think my math was off. If we're theoretically trying to see what clocks Drake's GPU would need to only be a 2x boost over Erista:
6 (6x the cores) / 3 = 2 (2x performance)
So we'd need a 3x reduction in clocks. Switch in handheld mode runs at ~307MHz at the lowest setting, so we'd need to see a 3x reduction of that, which would be ~102MHz. However, at least as far as Maxwell architecture is concerned this is impossible - the base clock frequency is ~76MHz, meaning every possible clock speed has to be a multiple of that base clock. So the lowest possible clock speed it could run at is ~152MHz, which would be about a 3x increase in GPU performance (due to the 6x increase in cores).
So 2x is completely off the table no matter the clocks, the absolute minimum is 3x.
(I have been using tildes because I do not feel like putting all the decimels for each frequency)
6 (6x the cores) / 3 = 2 (2x performance)
So we'd need a 3x reduction in clocks. Switch in handheld mode runs at ~307MHz at the lowest setting, so we'd need to see a 3x reduction of that, which would be ~102MHz. However, at least as far as Maxwell architecture is concerned this is impossible - the base clock frequency is ~76MHz, meaning every possible clock speed has to be a multiple of that base clock. So the lowest possible clock speed it could run at is ~152MHz, which would be about a 3x increase in GPU performance (due to the 6x increase in cores).
So 2x is completely off the table no matter the clocks, the absolute minimum is 3x.
(I have been using tildes because I do not feel like putting all the decimels for each frequency)
- Pronouns
- He/Him
I don’t quite understand why people are comparing teraflops of two different architecture‘s for PC platforms which are constantly evolving, I made a post several pages back detailing why comparing the teraflops of console to console cannot be compared to the teraflops as it exists in the PC space.
Games made for PC are practically zero, games made for consoles are infinitely more, reason is the console always gets unique optimization done for it which does not reflect the actual teraflop number that is being presented on paper.
If in the PC space a 10TFLOP card is performing equally to a 20TFLOP card, that’s because of a multitude of other reasons on an evolving platform. If it’s a 3TFLOP game console versus a 300 GFLOP game console, that’s a metric that can give you the idea of a 10x increase in the relative possible performance because developers were actually optimized for whatever that device is. However, in the sense here the increase in performance would be leveraged by how much relative hardware increase it already has beyond the TFLOP (256 vs 1536)
Consoles are very difficult to compare to a PC because they do not translate literally at all.
To give an example, the GTX 750 or 760 was compared to the PlayStation4 in relation to performance at launch in 2013, as time evolved it turns out that the most comparable graphics card in the PC space to the PlayStation4 was actually the GTX 1050 even though the PlayStation 4 GPU is ancient as Hell by the time that the GTX 1050 came out. The GTX 750/760 fell on the wayside ages ago and is noticeably weaker than the PlayStation4 in a practical sense.
Here’s another thing, graphics cards have different architecture, a graphics card with architecture a that has 10Tflop of FP32 and 20Tflop of FP 16 is performing equally to a graphics card from a different architecture that has 20Tflop of FP 32 and FP 16.
Why is that?
1) drivers
2) directx API version
3) game engines
4) tailored (or lack thereof) platforms of development.
Number three is the biggest point here, because if a game engine does not make for use of the graphics card and how it execute certain tasks, then it is a useless comparison. You can have ampere which is designed for a game engine of tomorrow, and you can have RDNA2 which can be designed for game engines of the past and how they were programmed for. How they perform now vs 5 years from now paints a different picture.
It’s a useless comparison using an evolving PC environment unlike a fixed platform like a console imo. They are not at all comparable.
Games made for PC are practically zero, games made for consoles are infinitely more, reason is the console always gets unique optimization done for it which does not reflect the actual teraflop number that is being presented on paper.
If in the PC space a 10TFLOP card is performing equally to a 20TFLOP card, that’s because of a multitude of other reasons on an evolving platform. If it’s a 3TFLOP game console versus a 300 GFLOP game console, that’s a metric that can give you the idea of a 10x increase in the relative possible performance because developers were actually optimized for whatever that device is. However, in the sense here the increase in performance would be leveraged by how much relative hardware increase it already has beyond the TFLOP (256 vs 1536)
Consoles are very difficult to compare to a PC because they do not translate literally at all.
To give an example, the GTX 750 or 760 was compared to the PlayStation4 in relation to performance at launch in 2013, as time evolved it turns out that the most comparable graphics card in the PC space to the PlayStation4 was actually the GTX 1050 even though the PlayStation 4 GPU is ancient as Hell by the time that the GTX 1050 came out. The GTX 750/760 fell on the wayside ages ago and is noticeably weaker than the PlayStation4 in a practical sense.
Here’s another thing, graphics cards have different architecture, a graphics card with architecture a that has 10Tflop of FP32 and 20Tflop of FP 16 is performing equally to a graphics card from a different architecture that has 20Tflop of FP 32 and FP 16.
Why is that?
1) drivers
2) directx API version
3) game engines
4) tailored (or lack thereof) platforms of development.
Number three is the biggest point here, because if a game engine does not make for use of the graphics card and how it execute certain tasks, then it is a useless comparison. You can have ampere which is designed for a game engine of tomorrow, and you can have RDNA2 which can be designed for game engines of the past and how they were programmed for. How they perform now vs 5 years from now paints a different picture.
It’s a useless comparison using an evolving PC environment unlike a fixed platform like a console imo. They are not at all comparable.
How? It quite literally has to be clocked below what the switch even does.
Spork4000
Tektite
- Pronouns
- He/Him
Why? Why do you think it's too high?
Angel Whispers
Follow me at @angelpagecdlmhq
Hello there, and thanks for writing. There isn't a relationship as such, BUT I had asked "What kind of games do you expect?", and it's going to be on PS5/XS/Switch, with support, one would imagine, over the course of a generation. The logic follows that if there is a Switch version, it, too, will be supported. If it is coming to the Switch's successor, rather than the existing Switch, which one might expect by the time it's ready, the logic follows that it would have to have enough about it to have the title in the first place. My point here is that if you're on the side of pessimism and high reservation, which is masquerading as "realistic" at this point, tbqh, there must be a coherent reasoning behind it, and "Because Nintendo" is no premise at all. The questions are there to help develop that reasoning, because I want to hear it, and have a better understanding of that point of view. I'm not even saying anybody is wrong here, but there is a thing called reasonable deduction, and I would hope that, compelled to think about these questions, one might conclude that we can surely expect more of Nintendo and their generational purpose, because the low end of expectation would represent failures to learn the lessons of the Wii U/3DS, and build on current success. I deliberately left little room for manoeuvre by sticking to the horse's mouth, using the maths to show what would need to happen to reach these very low points, and used the conditions of the 2017 Switch to show that even if you applied everything in the same way to the leaked SoC, you would still have something much better than what some are expecting... So, trying to "Because Nintendo" one's way into something which is demonstrably, highly improbable for the hell of it is unhealthy and stinks of bad faith discourse. This mindset among some fellow Nintendo fans needs to stop.
I have also said in the past that we're all adults here, so, the onus is on the individual to get a grip of their own emotions, rather than demand that everybody else Woo-hahs and keeps their expectations in check. If something better happens, what's next? That mindset doesn't allow you to enjoy the possibilities. If it doesn't happen, it doesn't happen, and that's OK, too... A few trolls might laugh, but so what? None of us are seeking their permission, I would hope, and it doesn't mean one should never dare to draw more optimistic conclusions, or even expect again in the future on their platform of choice.
- Pronouns
- He/Him
Basically, yeah. Or if they disable some SMs, which isn't likely considering the API only ever looks for 12.
- Pronouns
- She/They
On a lighter note, I had a funny thought recently: if RDR2 is still coming to Drake, it might be the only non-PC version to run at 60fps depending on how many sacrifices they're willing to make. Imagine going back in time and telling people that the best-performing version of that game would be a Switch port
"it's nintendo"
therefore Drake confirmed for 152MHz/456MHz! Nintendo fucking up!
maybe they canned the drake version alongside the current gen
Angel Whispers
Follow me at @angelpagecdlmhq
Except the point still stands because it shows that a sizeable amount of Switch owners have other platforms, and those numbers are enough to remind Nintendo that there is no room for complacency. I also added phones and tablets from 2018 at the earliest, then gaming laptops and PCs. The Wii U/3DS era should be enough to reinforce that, but even more this time around, because there is no 3DS to help them in the event of such another storm, and even that was a failure, relative to its predecessors. They also need the 3rdP sales to achieve/sustain similar levels of success, if not growth. "I buy Nintendo platforms to play their exclusives AS WELL AS 3rdP titles" is better than "I buy Nintendo platforms to play their exclusives" - Thankfully, Nintendo understands this, and that it needn't be a zero-sum question.
Angel Whispers
Follow me at @angelpagecdlmhq
LMAO, this made me chuckle because it's exactly how "Because Nintendo" logic sounds. The tragedy is that someone will write an article claiming he actually said it before some fans allow themselves to expect more.
Spork4000
Tektite
- Pronouns
- He/Him
I really think we need to put Nintendo's historically weaker hardware in context.
The Wii was a hail mary after the very powerful Gamecube.
The Wii U tried to absorb be Wii's architecture and hitched it's wagon to the second screen gimmick that developers were trying to make a happen at the time.
The 3DS had to support three screens at once, left eye, right eye and bottom, which was computationally expensive and a drain on their BOM. Not to mention them having to deal with Nvidia screwing them over by not meeting design specifications at the last second.
That is it, that is the entire history of Nintendo hardware being underpowered at the time of release.
Now assuming the Switch Successor only exists because devs want more power and it needs to be positioned favorably next to competition to keep third parties happen, then of course it'll be as powerful as it can be within certain design constraints, likely size and tdp. With all of that and the Nvidia leak there's no reason outside of pure pessimism to assume something as low as a 2x performance uplift. That would be one of the lowest generational leaps in the history of console gaming.
The Wii was a hail mary after the very powerful Gamecube.
The Wii U tried to absorb be Wii's architecture and hitched it's wagon to the second screen gimmick that developers were trying to make a happen at the time.
The 3DS had to support three screens at once, left eye, right eye and bottom, which was computationally expensive and a drain on their BOM. Not to mention them having to deal with Nvidia screwing them over by not meeting design specifications at the last second.
That is it, that is the entire history of Nintendo hardware being underpowered at the time of release.
Now assuming the Switch Successor only exists because devs want more power and it needs to be positioned favorably next to competition to keep third parties happen, then of course it'll be as powerful as it can be within certain design constraints, likely size and tdp. With all of that and the Nvidia leak there's no reason outside of pure pessimism to assume something as low as a 2x performance uplift. That would be one of the lowest generational leaps in the history of console gaming.
- Pronouns
- He/Him
For the record, for the Drake to be only 2x stronger in docked mode, it has to be clocked to around 384MHz. Half that for portable mode.
lol
There’s literally no point in turning the device on at that point.
lol
There’s literally no point in turning the device on at that point.
Thugstas
Bob-omb
Maybe they texted it back in 2018 with Xavier AGX/NX ?
LoneRanger
Chain Chomp
- Pronouns
- He/Him
The big if in next Switch HW for current gen ports is its CPU performance.
GPU raw power being the same as PS4 with DLSS, modern architecture, RT, etc would be really fine.
GPU raw power being the same as PS4 with DLSS, modern architecture, RT, etc would be really fine.
Z0m3le
Bob-omb
Yeah problem with low clocks is just that the GPU is so big, why use all those transistors on a big GPU to only get the performance of a GPU half it's size? The clocks for Switch also need to be hit for compatibility, so it's very unlikely that it is sub 2.36TFLOPs when docked.
Kopite7 who originally leaked T239's (Drake) existence by that model number and confirmed in the Nvidia leak, said that the CPU architecture is the same as Orin, so A78 cores, which performance per clock is better than Ryzen 2, so it just depends what it is clocked at, but 50-60% of PS5's CPU performance is very possible and should allow any current gen game to run on Drake well.
This game is a blurfest if you play it on PC on a non-4K monitor even with DLSS, hopefully R* figure this out on Drake.
Angel Whispers
Follow me at @angelpagecdlmhq
Some good points there. To add to the one about Nvidia not meeting the required specifications, I'm sure that's something they don't want to happen again, and the fact that they had continued to court Nintendo when they themselves failed would tell me they're determined more than ever to do their best to maintain their partnership. When Nintendo has something they consider a failure, the succeding generation usually sidesteps in terms of performance, and another approach is adopted. GameCube to Wii is the precedent, but Wii U/3DS to Switch did the same. When they're successful, the succeeding platform takes leaps - Wii to Wii U did this, while the portable line operated a generation or two behind the home line, until phones and tablets disrupted that. The move from 3DS to Switch was very much a leap compared to Wii U to Switch's small step, but mobile tech was advancing at a rapid rate, which meant that, whether Nintendo wants it or not, they have to beat phones and tablets. To paint a picture here, GBA as a portable SNES was awesome in 2001, Super Mario 64 on DS blown minds in 2004, but LOZ: Ocarina Of Time 3D on 3DS in 2011 did not have the same impact. The 3DS barely managed half the sales of its predecessor, but about 75m of anything is surely a success, right? Not here, it's a failure relative to its predecessors, which isn't talked about enough because the Wii U's fate was much worse. Another narrative which exists is that the 3DS still did well in the face of the rising popularity of phones and tablets... but that surely reinforces what I was saying about why delivering a definitive portable experience is imperative to success?
Perhaps, In another era, without phones and tablets, the 3DS might well have reached 100m, but the reason why it didn't have the same impact was that the iPad had launched a year earlier, while phones had better displays, and the masses had already seen what was possible on mobile/portable tech... on ARM-derivative devices, no less. There was an overwhelming sense that everybody had "moved on" from SD gaming, which, in turn, meant that Nintendo couldn't be complacent in a belief that they could run their portable line in the same way, and expect to copy and paste their past successes in this area. For all the successes that the 3DS mustered, the Vita, also a commercial failure, was able to receive 3rdP support in areas which the 3DS couldn't. So, What is the lesson to be learned from eleven years ago? After DS versions of Super Mario 64 and Diddy Kong Racing, "More N64 ports, but with bolted-on 3D effects" on their successor was a certain way to hurt their brand, and an overall decline, rather than hopeful growth, was the consequence. Does this sound familiar at all? Yet here we are, expecting that Nintendo and Nvidia, on the crest of the Switch's successful wave, won't deliver what is required to ensure they won't be so susceptible to that fate... One despairs.
Last edited:
- Pronouns
- He/Him
Kopite never said anything about the CPU architecture.
They only said, and I quote:
“This is a preliminary picture of T234 in Wikipedia. Very clear. So why do we always guess? Nintendo will use a customized one, T239.”
In fact, Kopite never discussed the CPU of any tegra ever.
It’s inferred based on the data breach that it uses A78, because ORIN uses it. But Kopite never explicitly said it was the A78, or implicitly. They only refer to the GPU elements and that’s it.
Last edited:
RDR2 is heavily filtered by design. they definitely won't fix it since it's working as intended
ShadowFox08
Paratroopa
What we know with the chip? If we are getting Drake, then +1500 GPU cores, and 102GB/s bandwidth are all but confirmed.
You said 2x GPU, RAM and bandwidth. 800 Maxwell TFLOPs, 8GB RAM and 52 GB/s vs Switch. Those are all pretty small jumps for a console with brand new hardware. The RAM and CPU are modest but not flat out awful.
If you meant handled mode being 2x as more powerful as switch docked, that makes sense. The GPU and bandwidth don't make sense.
Those specs are not gonna beat steam deck. Well maybe with DLSS if would be on par on some cases.
edit: I hope I didn't sound hostile
Last edited:
- Pronouns
- He/Him
yeah, RDR2 is one of the game examples I used across the thread that there’s a game with DLSS, and then there’s a game with proper and good DLSS.
DLSS is working as intended on RDR2, but it doesn’t look good actually. It’s by design of how the game is rendering, and DLSS exposes it more clearly what RAGE TAA was hiding very well because it’s very soft with it. RDR2 uses a checkerboard technique it seems for hair and leave and such. It was clearly designed for the consoles first which can hide it well enough.
DooM Eternal is another example of real proper and good DLSS implemented into the game, but it’s also Uber optimized to the point that you should wonder why use it if it runs so beautifully lol.
DLSS is working as intended on RDR2, but it doesn’t look good actually. It’s by design of how the game is rendering, and DLSS exposes it more clearly what RAGE TAA was hiding very well because it’s very soft with it. RDR2 uses a checkerboard technique it seems for hair and leave and such. It was clearly designed for the consoles first which can hide it well enough.
DooM Eternal is another example of real proper and good DLSS implemented into the game, but it’s also Uber optimized to the point that you should wonder why use it if it runs so beautifully lol.
LiC
Member
Saying that T239 is a customized T234 does imply that it also uses A78s (T234/T239 are names for the SoC, not the GPU only). Changing the CPU to a different design is a little bit more than customization.
Z0m3le
Bob-omb
Yeah even Alex (The PC expert) on DF recently said that he expects the new Switch GPU to be noticeably better than the Steam Deck's. I think key here is that while portable raw performance might be close, DLSS will secure portable mode and now that even the current Switch is doing a temporal upscaling in the same vain as DLSS/FSR2 with the release of Xenoblade 3, even in portable mode. I think we can put to bed the idea that Nintendo wouldn't use DLSS in portable mode.
Docked performance will be much higher, something the Steam Deck just doesn't do... Giving the Switch a huge advantage, in fact one of the reasons you would use such a large GPU is to use higher clocks than expected while docked, because if your performance target is PS4 in portable mode (before DLSS, similar to steam deck) then you'd need a large GPU since your portable clocks are limited, this means you can stay around 500MHz in portable mode (1.536TFLOPs) and have the room to increase the performance when docked, which you'd expect to be somewhere around double, or 1GHz (3.072TFLOPs)... This of course is if the process node is capable of those clocks within the TDP of this new model, generally I think they would need to move to 5nm (Samsung, or 7nm TSMC) to reach these sorts of speeds, given they are likely using 8 CPU cores this time, as the entire industry has made that a standard 9 years ago.
Yeah, I guess I implied that internally when I read it, generally speaking because he linked them, I linked their architectures, which has so far panned out, but technically Redd is right, reading that as if I was Drax the destroyer, I could absolutely separate the two chips from each other, natural implications do lean in T239 being a custom version of the T234 SoC, which in the real world makes a lot of sense as you can share R&D budget if they are sister chips instead of individually separate designs.
- Pronouns
- He/Him
They already have to change the CPU though, yes the A78AE is part of the A78 family of CPUs along with a cortex X1, A78 and the A78C, but you cannot simply swap it like a jigsaw piece, you have to quite literally redesign it for a different (but similar) CPU. A78AE has redundant extra logic that takes up more space on this silicon itself and each core itself is bigger than the regular A78.
If they were to use A78/C, they’d have to redesign it with less silicon space used in mind.
Unless they do use the A78AE which… I doubt.
LiC
Member
Shuffling something around for space reasons falls under customization. I don't know much about the electrical engineering side of a CPU, but I think it's reasonable to assume all A78 CPUs have mostly the same interfaces, and switching from one to the other would be much less work than plugging in a different type of CPU than the board was originally meant for -- probably no more work than the difference between Orin AGX and NX which needed to be rearranged for having different quantities of the same CPU core. Then you get to the driver and firmware side, which in Orin's case were all written for different core configurations of A78s and would need to be updated for a different CPU.
OP
OP
But just like how Nintendo can only decrease the frequencies so much before chip yields decrease, at the same time, Nintendo can only increase the frequencies so much before not only chip yields decrease, but also battery life decreases. Chip yields work both ways.
But saying that, I'm sure Nintendo and Nvidia are working to find a good balance between performance and battery life for handheld mode.
A hexa-core (6) configuration of the Cortex-A78C is just as probable as an octa-core (8) configuration of the Cortex-A78C.
102.4 GB/s is only confirmed as the max possible memory bandwidth, considering Drake has a max memory bus width of 128-bit, and assuming the LPDDR5 modules on Nintendo's new hardware runs at a max I/O rate of 6400 MT/s. And although I do think there's a possibility the LPDDR5 modules on Nintendo's new hardware runs at a max I/O rate of 6400 MT/s in TV mode, I also do think there's a possibility the LPDDR5 modules on Nintendo's new hardware is going to run at a reduced I/O rate in handheld mode (e.g. 5500 MT/s). The LPDDR4 modules on the Nintendo Switch ran at a max frequency of 1600 MHz when in TV mode, but ran at a reduced frequency of 1331 MHz when in handheld mode.
Last edited:
- Pronouns
- He/Him
D
Deleted member 887
Guest
I believe you misunderstand me. I'm not saying that Drake isn't intended for a Nintendo console. I'm 100% sure that it is. And I'm 95% sure that console is the long rumored 4k Switch revision.
However, 95% isn't 100%. Because (as I said before) this is not a position I believe, simply a scenario that is possible I'm going to put the whole analysis in a spoiler tag.
Orin and Drake are the only DLSS capable chips a new Switch could use that we know about and we only know details about Drake due to a hack. Consider what this conversation would be like if the Nvidia hack had not happened.
... a perfectly reasonable statement that we now know probably isn't true. Why is it impossible to believe that our new, hacked info still isn't 100% of the story either? We've connected hardware we know Nvidia is making for Nintendo (Drake) to the device that Nintendo is heavily rumored to be making (Switch Pro), but we also know that Nintendo is probably working a Next Gen device as well, at least conceptually. Why can't Drake be that? Why can't the Switch Pro be a SOC that wasn't leaked?
You said that the only viable chip for the DLSS Switch is Drake and... kinda? There have been persistent questions around how Nintendo is going to fit Dane quantities of silicon in the Switch form factor at acceptable power draw and heat levels, on this timeline, and why they would do it for a revision that mostly "just" plays Switch games? How much profit is possible on such a device? These open questions can be answered in one fell swoop with "it isn't."
The API name means little in this case. GX was the name of the GameCube API, and it was also the name of the Wii API. GX2 was the name of the WiiU API. NVN will be the base of Nintendo's internal graphics layer for the foreseeable future, as long as the internal architecture of their devices continues in this Vulcan-esque vein that Nvidia has been tapping for some time. If anything, the name NVN2 implies a next gen device and not a revision.
"Nintendo won't abandon their successful form factor" - probably not in the next 10 years, you are correct. But WiiU was both a continuation of AND a major revision of the value proposition of the Wii. It's not impossible Nintendo would try that direction again. It isn't impossible that they would try to develop a second line of hardware in much the same way they did with the DS and the Switch, where they launch a "third pillar" system, but then abandon or separate the lines of business if it's successful. Nintendo knows they can't ride the Switch forever, and that probably involves planning for a non-Switch system while the Switch still has enough gas in the tank that Nintendo can fall back to it if something goes wrong with their successor. It seems highly likely that those experiments would center around Switch-like internals (for ease of development, as well as ease of porting back to Switch, should something go wrong) that are too beefy to go in a handheld (to create a value proposition for a device that isn't a handheld.
TL;DR: It isn't logically inconsistent to believe in a Switch Pro and Drake without being certain they are the same device.
... a perfectly reasonable statement that we now know probably isn't true. Why is it impossible to believe that our new, hacked info still isn't 100% of the story either? We've connected hardware we know Nvidia is making for Nintendo (Drake) to the device that Nintendo is heavily rumored to be making (Switch Pro), but we also know that Nintendo is probably working a Next Gen device as well, at least conceptually. Why can't Drake be that? Why can't the Switch Pro be a SOC that wasn't leaked?
You said that the only viable chip for the DLSS Switch is Drake and... kinda? There have been persistent questions around how Nintendo is going to fit Dane quantities of silicon in the Switch form factor at acceptable power draw and heat levels, on this timeline, and why they would do it for a revision that mostly "just" plays Switch games? How much profit is possible on such a device? These open questions can be answered in one fell swoop with "it isn't."
The API name means little in this case. GX was the name of the GameCube API, and it was also the name of the Wii API. GX2 was the name of the WiiU API. NVN will be the base of Nintendo's internal graphics layer for the foreseeable future, as long as the internal architecture of their devices continues in this Vulcan-esque vein that Nvidia has been tapping for some time. If anything, the name NVN2 implies a next gen device and not a revision.
"Nintendo won't abandon their successful form factor" - probably not in the next 10 years, you are correct. But WiiU was both a continuation of AND a major revision of the value proposition of the Wii. It's not impossible Nintendo would try that direction again. It isn't impossible that they would try to develop a second line of hardware in much the same way they did with the DS and the Switch, where they launch a "third pillar" system, but then abandon or separate the lines of business if it's successful. Nintendo knows they can't ride the Switch forever, and that probably involves planning for a non-Switch system while the Switch still has enough gas in the tank that Nintendo can fall back to it if something goes wrong with their successor. It seems highly likely that those experiments would center around Switch-like internals (for ease of development, as well as ease of porting back to Switch, should something go wrong) that are too beefy to go in a handheld (to create a value proposition for a device that isn't a handheld.
TL;DR: It isn't logically inconsistent to believe in a Switch Pro and Drake without being certain they are the same device.
Again, I am not advocating this position but I'm not dismissing it either. This thread has a bad habit of being 90% confident on something and treating the 10% chance as impossible. This is especially annoying when the 10% chance proves right and we're accused of being idiots who make stuff up because we got it "wrong"
An unknown thing here, is of course to what degree next gen engines will rely on fast SSD when crossgen ends, what kind of storage Drake is packing, how much the cpu disparity will end up mattering post crossgen.
I predict the gpu will be the least of Drakes problems, when it comes to getting third party support.
coming off of Spider-man PC, I don't think the fast SSD will be as much of a problem as decompressing data will be. either dedicating a whole core or two to it, making custom hardware, or using gpu resources for
Angel Whispers
Follow me at @angelpagecdlmhq
Sure, I'm not suggesting that they'll clock it very highly in portable mode, but rather pointing out how relatively low it has to clock to achieve a certain level of performance, and that a relatively modest clock would eclipse both the Steam Deck and PS4's performance. In docked mode, you're not using the battery, so, the capacity to clock higher surely comes into play. I don't think the 6-core option is plausible for a few reasons. One is that octa-core processors have been standardised over the last decade in console game development, so, the logic follows that they'll adopt it now. Also, the 2017 Switch had 8 cores (2 clusters of 4), which couldn't be used together - One would imagine that had it been possible to do so, they would've done it, but the less performant cores had to be shut-off. Imagine, pre-Switch release, had we been speculating on that SoC - You would have people suggesting that "NX" would have a quad-core A53 Class CPU "Because Nintendo". They aren't averse to the idea of higher system performance. It is an option in the purest definition, but in terms of ambitions, it's not realistic. Another is that there is an Orin-class SoC with a weaker GPU (1024 CUDA Cores) than Codename Drake's and it has an octa-core processor. I strongly suspect that Nvidia wouldn't potentially bottleneck Nintendo's SoC design by pairing the stronger GPU with a hexa-core CPU, and ultimately failing to provide what developers have become used to for quite a while.
Spider Man is a ps4 game at heart?
- Pronouns
- He/Him
I’m not so sure about that, now I don’t have a PlayStation 5 at the moment, however I’m pretty sure that Spider-Man remastered is also released on the PlayStation 4, so game engines are not really tailored to take full advantage of an SSD this early on.
Blue Monty
Inkling
- Pronouns
- he/him
Spider-Man remasters didn’t release on the PS4 , the original game did. They remastered the game + made Miles Morales for PS5 only
OP
OP
The GPU's been the least of the Nintendo Switch's problems when third party support is concerned as well. The main bottlenecks for the Nintendo Switch are the CPU and RAM.
The CPU could still be a bottleneck for Nintendo's new hardware, albeit not as much as with the Nintendo Switch, especially if developers plan on releasing more open world games for the current gen consoles.
RAM could also still be a bottleneck for Nintendo's new hardware as well, although I don't know how Ampere compares to RDNA 2 when memory latency's concerned when video game consoles are concerned, especially with Infinity Cache being absent on the PlayStation 5's and Xbox Series X|S's GPUs.
But as mentioned, I can see internal flash storage being a possible major bottleneck if more current gen games take advantage of the faster sequential read/write speeds and higher random read/write performance offered by the internal NVMe SSDs on the PlayStation 5 and Xbox Series X|S, depending on what Nintendo chooses for the internal flash storage.
There also exists Jetson Orin NX (8 GB), which has 1024 CUDA cores, and 6 Cortex-A78AE cores.
Look over there
Bob-omb
I just want to point out that for the memory bandwidth, likely's probably at 4x.
IIRC, what is explicitly stated is that T239/Drake has 1 frame buffer partition, while T234/Orin has 2 frame buffer partitions, and that memory type is LPDDR5.
What we can then reasonably infer is that since Orin has 256-bit memory bus width, Drake would then be 128-bit wide. And that's double the bus width of base Switch. The type is most likely LPDDR5, though I am hoping for that outside shot of 5X. But anyway, 5 doubles the bandwidth of 4. Combine the two doubles and we get a probable 4x.
Going by TechPowerUp's database, ehhhh, yea. The 1080 averages out to 116% against the 3050's 100%. Looking at the specs to see what can explain that...
They're both 20 SMs big. Their clocks are pretty similar; the 1080's base clocks are a few percent higher while the 3050's boost clocks are a few percent higher. The major difference is in bandwidth; the 1080 can have either 320 or 352 GB/s while the 3050 has 224 GB/s. So it's arguable that the 3050 is bandwidth constrained compared to the 1080. For that matter, the 3050 also looks bandwidth constrained relative to the rest of the desktop RTX 30 series. For a quick and dirty check within the same generation, do bandwidth/tflops (the idea's really supposed to be more like bandwidth/(SM_count*clock), but within the same architecture, tflop can work as a shortcut for SM_count*clock) ; the 3050's result is noticeably behind the rest of the 30 series.
they changed how assets are handled to take advantage of the new decompression in PS5 and that's what was ported to the PC
Spiderman specifically was
People ITT still think games need a bajillion CPU cores and/or threads to scale?
A modern game typically has multiple systems each of which run on it's own software thread. Eg: physics, game logic, etc...
A single CPU core/thread can run hundreds if not thousands of software threads an application creates.
IPC and cpu cache size matters more if you know what unity and unreal engine teams have been doing to try and optimize their software to avoid issues like cache misses or games running poorly on the now old and moldy ivy/sandy bridge and AMD fx era pcs for example.
Drake could have 4 cores, wouldn't matter much if it comes with enough performance-per-core + 8GB or more of system RAM.
A modern game typically has multiple systems each of which run on it's own software thread. Eg: physics, game logic, etc...
A single CPU core/thread can run hundreds if not thousands of software threads an application creates.
IPC and cpu cache size matters more if you know what unity and unreal engine teams have been doing to try and optimize their software to avoid issues like cache misses or games running poorly on the now old and moldy ivy/sandy bridge and AMD fx era pcs for example.
Drake could have 4 cores, wouldn't matter much if it comes with enough performance-per-core + 8GB or more of system RAM.
Dekuman
Kremling
- Pronouns
- He/Him/His
Assuming this device is truly going to launch with all 12 SMs, what are your thoughts they went that way to make a Pro model possible on the 1st die shrink and either clock low on the OG model or turn off a couple of SMs for thermals , with the long shot option of unlocking them in a future patch on docked. Like a Boost mode of sorts. Either way, the SoC seems over engineered for a 4-5x leap from Switch we were expection. 12 SMs is 3x the low end expectations of 4SM a year ago, and double 6SM many expected. So easily this chip could be (in docked mode) 10x Switch docked with decent clocks and thermals.
Last edited:
ShadowFox08
Paratroopa
I'm not expecting half of it though. 88-102GB should definitely be the range, but considering they never mentioned 88GB, 102 (max speed for LPDDR5) should kinda be reasonably expected in docked mode. 88GB or lower for handheld, also makes sense.
Mr Swine
Like Like
- Pronouns
- He
How much space would a dual core A55 (OS only) take up at 8nm if Nintendo decides to go with Hexa-core A78/78C instead of a Octa-core variant? Would it be smaller or roughly the same size?
if they went with the A78C variant, there will be no little cores at all. in any case, it'd be a small amount of space, but that's still not really worth it, IMO
you'll need speed to overcome fewer cores and there's only so much power they can pump through the SoC
Look over there
Bob-omb
Theoretically, easily smaller. You can most likely fit 3 or 4 A55s in the same space as a A78. You get pretty close to 4 A55s within the area of 1 A75, and I don't think that whatever size difference there is between A75 and A78 is enough to significantly alter the math for the question here.
...of course, ARM never said anything about being able to use A55s with the A78C. The explicit examples they provided were hexa-core and octa-core A78C, with no A55s.
Edit: to flesh things out a bit more/reminder for the readers, about half a year after the A78C announcement (late 2020, so now I'm referring to ~May 2021), ARM did their regular yearly announcement. That year was the big update of everything to ARM v9. The regular DynamIQ Shared Unit got updated to allow for up to 8 big cores at once. It should be theoretically possible to put together a cluster of 6 A710s/A715 cores and 2 A510 complexes*.
*why 'complexes' instead of 'cores'? Just... read this overview.
...of course, ARM never said anything about being able to use A55s with the A78C. The explicit examples they provided were hexa-core and octa-core A78C, with no A55s.
Edit: to flesh things out a bit more/reminder for the readers, about half a year after the A78C announcement (late 2020, so now I'm referring to ~May 2021), ARM did their regular yearly announcement. That year was the big update of everything to ARM v9. The regular DynamIQ Shared Unit got updated to allow for up to 8 big cores at once. It should be theoretically possible to put together a cluster of 6 A710s/A715 cores and 2 A510 complexes*.
*why 'complexes' instead of 'cores'? Just... read this overview.
Last edited:
OP
OP
I don't think I mentioned the possibility of Nintendo reducing the I/O rate and memory bandwidth by half for handheld mode.
Also, I believe the Nvidia leaks never made any explicit mentions of the memory bandwidth, but rather the framebuffer, which leads to the max memory bus width. And yes, Nvidia never mentioned any Jetson Orin modules with a max memory bandwidth of 88 GB/s, but the Jetson Orin modules are not necessarily designed to be used in projects that necessitated great battery life.
As I've mentioned before, I think there's a possibility the LPDDR5 modules on Nintendo's new hardware runs at a max I/O rate of 6400 MT/s in TV mode, and at a reduced I/O rate in handheld mode.
6 Cortex-A78 cores (not the hexa-core variant of the Cortex-A78C) requires 2 CPU clusters since there can only be a max of 4 CPU cores per cluster for the Cortex-A78 (as opposed to a max of 8 CPU cores per cluster for the Cortex-A78C). And having more than 1 CPU cluster can introduce additional, unwanted latency since there are more hardware components that needs to be communicate with.
Hmm, I really don't follow here. I've been saying from the start that using full software emulation for Base Switch backwards compatibility is not feasable. You said earlier that a hardware solution would be more costly and pointless.
If it's neither software nor hardware, what would it be?
You talked about translation layers but wouldn't that just mean it's running natively and the incompatible API calls would be translated into equivalent compatible calls?
I think I just misunderstood what you said
- Pronouns
- He/Him
They sorta did!
Fun fact:
Here's how Arm's latest CPU targets laptop and handheld console performance
Arm's latest Cortex-A78C CPU is built to offer always-connected laptops and handheld console class performance.
www.androidauthority.com
So, A55 cannot be paired with the A78C 6 core or 8 core setup.
But theoretically, X1C can be paired with the A78C 6 or 8 core setup in a unique configuration that is 4+4 or 2+6 or 6+2.
And it would be smaller. 4 A55 should be similar to 1 A78.
Drake will be in a similar situation as PS5/XS, with a CPU compatible with its predecessor, but an incompatible GPU. Some light emulation will most likely be required to make all the Maxwell shaders shipped with Switch games run on the new Ampere GPU.
"NINTENDO DIRECT" FOR THE NEXT FEW MONTHS?
BY 2022, NINTENDO WOULD ALREADY HAVE FULL ANNOUNCEMENTS FOR THE VIDEO GAME CATALOG TO SELL.
PRECENDENTS:
WILL NINTENDO BOOST SALES BEFORE THANKSGIVING DAY OR CHRISTMAS WITH A NINTENDO DIRECT?
WHEN WILL THE "MARIO MOVIE" ADVERTISING BEGIN?
Analyzing the precedents, and if there is a "Nintendo Direct" in what remains of 2022, the dates for the announced could be: 22 August 2022, 10 October 2022, or the most likely 19 December 2022.
OPEN QUESTIONS:
BY 2022, NINTENDO WOULD ALREADY HAVE FULL ANNOUNCEMENTS FOR THE VIDEO GAME CATALOG TO SELL.
PRECENDENTS:
- NINTENDO WOULD BE GIVING A MONTH BETWEEN EACH TRIPLE TITLE RELEASED.
- In the absence of E3, Nintendo launched its own events such as: Nintendo Live 2022 (Japan, 8 and 9 October 2022) and Nintendo Switch Road Trip 2022 (USA, ends 5 September 2022).
- A World Cup that would freeze dates for a new event, from November 21 to December 18, 2022.
WILL NINTENDO BOOST SALES BEFORE THANKSGIVING DAY OR CHRISTMAS WITH A NINTENDO DIRECT?
WHEN WILL THE "MARIO MOVIE" ADVERTISING BEGIN?
Analyzing the precedents, and if there is a "Nintendo Direct" in what remains of 2022, the dates for the announced could be: 22 August 2022, 10 October 2022, or the most likely 19 December 2022.
OPEN QUESTIONS:
- WILL THE CAMPAIGN FOR XENOBLADE OR OTHER NEW FRANCHISES CONTINUE STRONG?
- WOULD BOTW2 BE THE FIRST GAME IN THE SAGA NOT TO BE SHOWN PLAYFULLY IN AN E3 BEFORE ITS RELEASE?
- 2015 NVIDIA SHIELD TV WITH TEGRA X1, 2019 TEGRA X1+ IMPLEMENTED ON NINTENDO SWITCH, WILL 2023 UPDATE/CHANGE FOR NINTENDO TEGRA?
- Pronouns
- He/Him
wait, what? how?
Last edited:
