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

[Discostew]

Something I'm surprised I don't recall seeing brought up before, when it comes to speculation on an OLED Lite or similar revision launching instead of an upgrade: the Lite and OLED both showed up in firmware at least 17-18 months before launch (15 months before announcement). Unless Nintendo suddenly started really caring about hiding revisions, which I doubt, and did so successfully, there's no way anything else is launching besides the eventual upgrade.

Did Switch data show up in past system firmware? Wii U? 3DS?

 

[ILikeFeet]

Did Switch data show up in past system firmware? Wii U? 3DS?

no. these are different OSs (despite the Horizon name also being the 3DS's OS) so there's no reason for Switch to show up there

 

[LiC]

Did Switch data show up in past system firmware? Wii U? 3DS?

No.

The reason why revisions showed up in Switch firmware, and a "new Switch"/successor/whatever wouldn't, is because the different Switch models all share the same firmware. The thing that gave away the revisions was configuration profiles which make that sole firmware behave differently depending on the type of device it's currently running on. This is why the OLED supporting a different kind of screen, docked display controller, etc. was known way ahead of time -- even though the other Switch models don't have those, support for them was loaded onto all models and then controlled with the profile selection.

A new system would have a separate firmware build entirely, so references to it will never appear in current Switch model firmware.

 

[Serif]

I wanna see the new startup animation for Switch Super

 

[Deleted member 887]

So

1) I FINISHED THE DLSS CALCULATOR
(and it's not looking that good...)

Here it is : https://www.calconic.com/calculator...or-3000/63dec5fcd47bb900296c3ff9?layouts=true

https://www.calconic.com/calculator...or-3000/63dec5fcd47bb900296c3ff9?layouts=true
Your numbers are identical to mine - not surprising because except for the sparsity boost, our methodology is identical.

However, I pulled my stuff because, like you, I didn’t have a lot of data points. There are some additional benchmarks out in the wild but I haven’t audited them for quality.

 

[Spheromancer]

My super official direct predictions:


Opening:

Shows off the Metroid Prime remake whatever it is, trilogy or just prime 1, releasing fall 2023

Mid portions:

Kirby's Dreamland Deluxe breakdown

3rd Parties

Hollow Knight Silksong June/July Release Date

Advanced Wars, releasing April 2023

Switch price cuts. OLED $300, OG $250, Lite $150

Pikmin 4 gameplay, November release date

Ending Portions:

Tears of the Kingdom full 3 minute reveal trailer, voice acting, story, locations etc. Short Aonuma segment after talking about the game. Still releasing 5/12, OLED reveal/pre orders.

One last thing: Metroid Prime 4 cinematic teaser, 2024 release date





Obviously some other stuff/xenoblade/bayo/mario movie/dlc will be there somewhere but idc about them so I didnt write it

​

 

[Gerald]

What do you think about design? Will be some changes? imo maybe will be some revolution with Joy-Con

 

[chocolate_supra]

My super official direct predictions:


Opening:

Shows off the Metroid Prime remake whatever it is, trilogy or just prime 1, releasing fall 2023

Mid portions:

Kirby's Dreamland Deluxe breakdown

3rd Parties

Hollow Knight Silksong June/July Release Date

Advanced Wars, releasing April 2023

Switch price cuts. OLED $300, OG $250, Lite $150

Pikmin 4 gameplay, November release date

Ending Portions:

Tears of the Kingdom full 3 minute reveal trailer, voice acting, story, locations etc. Short Aonuma segment after talking about the game. Still releasing 5/12, OLED reveal/pre orders.

One last thing: Metroid Prime 4 cinematic teaser, 2024 release date





Obviously some other stuff/xenoblade/bayo/mario movie/dlc will be there somewhere but idc about them so I didnt write it

​

Did you mean for this to go in the Direct thread?

 

[Concernt]

No.

The reason why revisions showed up in Switch firmware, and a "new Switch"/successor/whatever wouldn't, is because the different Switch models all share the same firmware. The thing that gave away the revisions was configuration profiles which make that sole firmware behave differently depending on the type of device it's currently running on. This is why the OLED supporting a different kind of screen, docked display controller, etc. was known way ahead of time -- even though the other Switch models don't have those, support for them was loaded onto all models and then controlled with the profile selection.

A new system would have a separate firmware build entirely, so references to it will never appear in current Switch model firmware.

"Never appear"?

Come on, be serious, parts of it already have. They're definitely being more clean about it than OLED Model but people have been daramining stuff about a new model after OLED Model for months.

 

[Concernt]

What do you think about design? Will be some changes? imo maybe will be some revolution with Joy-Con

I severely doubt there will be changes other than the backplate and cooling system.

 

[Concernt]

Did you mean for this to go in the Direct thread?

Technically the OLED Model reveal IS hardware speculation.

 

[LiC]

"Never appear"?

Come on, be serious, parts of it already have. They're definitely being more clean about it than OLED Model but people have been daramining stuff about a new model after OLED Model for months.

Parts of it? Such as?

The only things that have appeared are loose hints which we can interpret to indicate the kernel may have undergone development to support another platform. What I said is that there are no references to another system, and there never will be.

 

[JoshuaJSlone]

So

1) I FINISHED THE DLSS CALCULATOR
(and it's not looking that good...)

Here it is : https://www.calconic.com/calculator...or-3000/63dec5fcd47bb900296c3ff9?layouts=true

Obviously you're way way more knowledgeable about this stuff than I am, but it seems weird to me that when comparing the time costs of DLSS Performance in 4K and 1440p, the pixel difference for 4K is +125%, but the time cost is +187%? Is it possible to explain like I'm five? It seems counterintuitive that having a higher render resolution would slow the DLSS calculations down so significantly, but maybe that's the case.

 

[Aether]

https://www.calconic.com/calculator...or-3000/63dec5fcd47bb900296c3ff9?layouts=true
Your numbers are identical to mine - not surprising because except for the sparsity boost, our methodology is identical.

However, I pulled my stuff because, like you, I didn’t have a lot of data points. There are some additional benchmarks out in the wild but I haven’t audited them for quality.

im kinda trying to interpret what you guys are trying to say.
The frame timings on the proposed GPU with some expected clocks would be strict?
Yeah, they would, but that's also depending on what and how its rendered, games created with DLSS in mind would need to take that into account. Its not as if you would develop a game with the same mindset as games that don't need to calculate DLSS frames?

 

[Paul_Subsonic]

Obviously you're way way more knowledgeable about this stuff than I am, but it seems weird to me that when comparing the time costs of DLSS Performance in 4K and 1440p, the pixel difference for 4K is +125%, but the time cost is +187%? Is it possible to explain like I'm five? It seems counterintuitive that having a higher render resolution would slow the DLSS calculations down so significantly, but maybe that's the case.

The answer is really simple : I have no fucking clue. Also according to DF data, 1080p is basically identical to 1440p (a 10% difference).
You'll note that in the calculator, besides the 1440p section there's an information section where I say I don't trust 1440p results.
The reason why I trust 4K over 1440p is because the 4k measurements were done on 2 seperate ampere cards, but the 1440p only on one 2060, of Turing architecture (so a different one).
The 4K results are to be taken with a very big pinch of salt, the 1440p results with a truckload of it.

 

[Hermii]

The answer is really simple : I have no fucking clue. Also according to DF data, 1080p is basically identical to 1440p (a 10% difference).
You'll note that in the calculator, besides the 1440p section there's an information section where I say I don't trust 1440p results.
The reason why I trust 4K over 1440p is because the 4k measurements were done on 2 seperate ampere cards, but the 1440p only on one 2060, of Turing architecture (so a different one).
The 4K results are to be taken with a very big pinch of salt, the 1440p results with a truckload of it.

In addition to that, it should also be slightly faster on Drake due to being tightly integrated into nvn. Hard to estimate how much that would speed things up.

 

[Dekuman]

Chaos theory: Pikmin takes the May slot and TOTK is delayed into fall 2023 or early 2024. ;P

 

[Paul_Subsonic]

im kinda trying to interpret what you guys are trying to say.
The frame timings on the proposed GPU with some expected clocks would be strict?
Yeah, they would, but that's also depending on what and how its rendered, games created with DLSS in mind would need to take that into account. Its not as if you would develop a game with the same mindset as games that don't need to calculate DLSS frames?

I'm not sure I fully understand the question, so I'm gonna answer the best I can.

A 30fps frame is always gonna be 33.3 ms. Because you've gotta have 30 of them, per second. The idea behind the calculator is how long does DLSS take in that budget. Because that budget will be split in 2 : the DLSS and the actual rendering (to simplify). Obviously, with a lower ms budget, you can do less stuff.

What this means is that you can't allocate too much time to DLSS, or else you won't have enough time to do rendering. Devs will take that into account, but they won't want to sacrifice too much for DLSS. So, for 30 fps, if DLSS takes 5ms it's fine cause it leaves 28 ms for rendering. But if DLSS takes 15ms, it only leaves 18ms for rendering, which will means significant sacrifices in graphics - sacrifices that devs will probably not make.

 

[Concernt]

Parts of it? Such as?

The only things that have appeared are loose hints which we can interpret to indicate the kernel may have undergone development to support another platform. What I said is that there are no references to another system, and there never will be.

Semantics- literally.

 

[LiC]

Semantics- literally.

No, there is a giant substantive difference between seeing "Aula, a new Switch model using the Mariko SoC and 4 GB of LPDDR4X which supports a new screen driver and has a specific Realtek display chip in the dock" and "a gap in system call IDs which may or may not be used by some other hardware."

 

[Pokemaniac]

Chaos theory: Pikmin takes the May slot and TOTK is delayed into fall 2023 or early 2024. ;P

Counterpoint: Drake launches with Zelda this may, then Pikmin comes shortly afterwards as a big hardware showcase.

Probably not happening unless this year is absolutely packed, but it's an interesting idea.

 

[Aether]

I'm not sure I fully understand the question, so I'm gonna answer the best I can.

A 30fps frame is always gonna be 33.3 ms. Because you've gotta have 30 of them, per second. The idea behind the calculator is how long does DLSS take in that budget. Because that budget will be split in 2 : the DLSS and the actual rendering (to simplify). Obviously, with a lower ms budget, you can do less stuff.

What this means is that you can't allocate too much time to DLSS, or else you won't have enough time to do rendering. Devs will take that into account, but they won't want to sacrifice too much for DLSS. So, for 30 fps, if DLSS takes 5ms it's fine cause it leaves 28 ms for rendering. But if DLSS takes 15ms, it only leaves 18ms for rendering, which will means significant sacrifices in graphics - sacrifices that devs will probably not make.

...yeah. DLSS is a trade of, we knew as much. Like literary every graphical feature. they can decide to do expensive multi pass renderings, do some complex filtering, invest the time in shadows and lighting, or to render a rather basic low resolution version and pump that up to 4k using dlss. It was always a trade off, and will always be, and depending on genre, game, artstyle, the trade offs will favor one aspect of the render pipeline over the other.

I never bought the "DLSS" is magic, precisely because of that.
Thats why im confused about those calculations, without any context (what how and why stuff is rendered in a game, type of game, etc) they are kinda pointless.

 

[Deleted member 887]

im kinda trying to interpret what you guys are trying to say.
The frame timings on the proposed GPU with some expected clocks would be strict?
Yeah, they would, but that's also depending on what and how its rendered, games created with DLSS in mind would need to take that into account.

DLSS 2 is a way of using AI to draw pixels, rather than having your engine draw pixels. If DLSS 2 is slower than your engine, then DLSS 2 isn't an advantage.

Lemme do some simple math.

DLSS 2 Quality Mode doubles the image resolution. Another way to double the image resolution? Spend twice as much time drawing it.

A 60fps game has only 16.6ms to draw a frame - one sixtieth of a second. So the break even point of DLSS 2 Quality Mode is 8ms. Any longer than that, then you can just draw the pixels yourself.

Except not all of frame time is spent drawing a frame. Usually about a third is spent doing CPU stuff - enemy AI, physics calculations, playing audio, reading player input. So really you only have 5ms before Quality Mode is unusable.

Its not as if you would develop a game with the same mindset as games that don't need to calculate DLSS frames?

Literally every game currently using DLSS was developed without DLSS in mind.

 

[Deleted member 887]

Thats why im confused about those calculations, without any context (what how and why stuff is rendered in a game, type of game, etc) they are kinda pointless.

Yeah, to add additional context to my previous post - it's not a trade off like other graphical tradeoffs. It is an alternate way to draw pixels, not a technique to make pixels prettier.

If you have 1 million pixels on screen, you have to draw 1 million pixels. Most graphical tradeoffs - shadow complexity, level of detail, texture resolution - are about how much detail is in those 1 million pixels that you draw.

Upscaling techniques are about drawing only some of those 1 million pixels and letting the upscaler draw the rest faster than you could do it natively. When the scaling factor - the percentage of the pixels the upscaler draws - is lower than the upscaler's cost - the percentage of frame time the upscaler takes to work - then it is always a loss. 100% of the time.

 

[LiC]

DLSS 2 is a way of using AI to draw pixels, rather than having your engine draw pixels. If DLSS 2 is slower than your engine, then DLSS 2 isn't an advantage.

Lemme do some simple math.

DLSS 2 Quality Mode doubles the image resolution. Another way to double the image resolution? Spend twice as much time drawing it.

A 60fps game has only 16.6ms to draw a frame - one sixtieth of a second. So the break even point of DLSS 2 Quality Mode is 8ms. Any longer than that, then you can just draw the pixels yourself.

Except not all of frame time is spent drawing a frame. Usually about a third is spent doing CPU stuff - enemy AI, physics calculations, playing audio, reading player input. So really you only have 5ms before Quality Mode is unusable.


Literally every game currently using DLSS was developed without DLSS in mind.

Have all these comparisons been using Quality mode? I haven't really looked. Because Performance mode is more likely what developers will tend to use on Nintendo's next hardware.

Also, as for the frame time thing, the ideal is for CPU and GPU work to happen entirely in parallel, not sequentially. This limit isn't achievable in practice, but developers attempt to minimize the time where either the CPU or GPU is idle. So I wouldn't say there's some fixed number of milliseconds that has to be "CPU time" only.

 

[Aether]

DLSS 2 is a way of using AI to draw pixels, rather than having your engine draw pixels. If DLSS 2 is slower than your engine, then DLSS 2 isn't an advantage.

Lemme do some simple math.

DLSS 2 Quality Mode doubles the image resolution. Another way to double the image resolution? Spend twice as much time drawing it.

A 60fps game has only 16.6ms to draw a frame - one sixtieth of a second. So the break even point of DLSS 2 Quality Mode is 8ms. Any longer than that, then you can just draw the pixels yourself.

Except not all of frame time is spent drawing a frame. Usually about a third is spent doing CPU stuff - enemy AI, physics calculations, playing audio, reading player input. So really you only have 5ms before Quality Mode is unusable.

I get that, but i seemingly missed it, was that a comparison between native 1440 and uprezzed to 1440?
Then i missread the actual proposal there.


... oh, i think now i understand the comparison.
Yeah, but that kinda assumes that all aspects of the rendering pipeline scale the same in regards to compute power, doesn't it?
And depending on the testing they then decide to render this in half frames, this in a lower resolution,etc.
Meaning just rendering in double the time is not the same.
Maybe im just miss judging some aspects there.

Literally every game currently using DLSS was developed without DLSS in mind.

And how many of those are made under the assumption that the platform 100% has DLSS? none. On pc DLSS is a nice to have and cant be expected. On switch 2 this would be different, they know exactly its there and what its limitations are. The benefit of a known target.

Yeah, to add additional context to my previous post - it's not a trade off like other graphical tradeoffs. It is an alternate way to draw pixels, not a technique to make pixels prettier.

If you have 1 million pixels on screen, you have to draw 1 million pixels. Most graphical tradeoffs - shadow complexity, level of detail, texture resolution - are about how much detail is in those 1 million pixels that you draw.

Upscaling techniques are about drawing only some of those 1 million pixels and letting the upscaler draw the rest faster than you could do it natively. When the scaling factor - the percentage of the pixels the upscaler draws - is lower than the upscaler's cost - the percentage of frame time the upscaler takes to work - then it is always a loss. 100% of the time.

Sure (i know that DLSS works on finished frames and motion vectors), but isnt the benefit of it that its a really capable uprenderer?
If i just draw the pixely myself, then i have to take care how the lower resolution renders (shadows for example) are interpolated, wouldnt DLSS be a really great finished solutio for that problem, instead of doing my own work?
Then there is also the fact that it uses the tensor cores, if i do it, i have to do that on the cores i could start doing the next frame? or are we assuming that the tensor cores are being used for regular graphics computations if no DLSS is being used?

 

[Deleted member 887]

Have all these comparisons been using Quality mode? I haven't really looked. Because Performance mode is more likely what developers will tend to use on Nintendo's next hardware.

I suspect we'll see ultra performance mode as well for 4K output. The struggle is that DLSS's performance curve scales (mostly) with output resolution, not scaling factor. So 4K output costs the same (almost) regardless of what base resolution you start with.

Also, as for the frame time thing, the ideal is for CPU and GPU work to happen entirely in parallel, not sequentially. This limit isn't achievable in practice, but developers attempt to minimize the time where either the CPU or GPU is idle. So I wouldn't say there's some fixed number of milliseconds that has to be "CPU time" only.

Absolutely true. I took the ~5ms numbers from the GDC data a few years back about post-processing effects and frame times. But yes, it is variable. There is that constant tug between "explaining why these numbers are eyebrow raising" and "laying down all the nuance."

I think @Paul_Subsonic's analysis is solid, but I can't imagine enabling hardware accelerated DLSS on hardware where it is effectively unusable. So I'm pointing in the direction of "very small number of usable benchmarks" and shrugging my shoulders until there is more data to go on.

 

[Deleted member 887]

And how many of those are made under the assumption that the platform 100% has DLSS? none. On pc DLSS is a nice to have and cant be expected. On switch 2 this would be different, they know exactly its there and what its limitations are. The benefit of a known target.

I don't think that actually holds up when frame times are huge - but even if it did, you're ruling out ports being able to use DLSS 2 to squeeze games onto the Switch without having to rewrite the rendering pipelines from the ground up for the hardware.

Sure (i know that DLSS works on finished frames and motion vectors), but isnt the benefit of it that its a really capable uprenderer?

Yeah, it's extremely capable. Almost as good as drawing the pixels myself.

If i just draw the pixely myself, then i have to take care how the lower resolution renders (shadows for example) are interpolated, wouldnt DLSS be a really great finished solutio for that problem, instead of doing my own work?

Not really. You have to do all that regardless, you need a complete rendering pipeline that generates a totally complete image before DLSS can function.

DLSS 2 has one advantage over native rendering, which is subpixel stability - the fizzing that can happen when lots of highly detailed objects, like foliage, move on screen. But that's not a clear win when the rest of image quality is coming down.

Then there is also the fact that it uses the tensor cores, if i do it, i have to do that on the cores i could start doing the next frame? or are we assuming that the tensor cores are being used for regular graphics computations if no DLSS is being used?

DLSS 2 still runs on the GPU as a whole, it's just that it's core uses special instructions the tensor cores make fast. You can overlay CPU and GPU work like this - in fact, that's how DLSS 3 works - but I'm not sure you can overlay Tensor Core and GPU work like this in any way that is a significant win.

 

[Aether]

Absolutely true. I took the ~5ms numbers from the GDC data a few years back about post-processing effects and frame times. But yes, it is variable. There is that constant tug between "explaining why these numbers are eyebrow raising" and "laying down all the nuance."

I think @Paul_Subsonic's analysis is solid, but I can't imagine enabling hardware accelerated DLSS on hardware where it is effectively unusable. So I'm pointing in the direction of "very small number of usable benchmarks" and shrugging my shoulders until there is more data to go on.

That kinda... Its a complex prozess for a frame from start to finish. Some aspects can be done in parallel, some less so, and i just feel like looking at it in that strict of a way (if it takes longer then x its useless) is less usable to say a system where if is tacked on by design (pc), at least before we have more examples/data points from the first games running on that system with dlss.


In my personal opinion 4k will be an outlier for less complex games and most will target 1080-1440p

 

[Gerald]

I have a feeling that Nvidia itself wants the Switch 2 to be a powerful console

 

[Aether]

I don't think that actually holds up when frame times are huge - but even if it did, you're ruling out ports being able to use DLSS 2 to squeeze games onto the Switch without having to rewrite the rendering pipelines from the ground up for the hardware.

Thats true, in that case lowering the resulution really low (say 600p) wouldbe an option, or having 720 rendered to 1080?
I dont expect games that where build for a completly different archtitecture/constelation to be easy to port in high resolution. Switch 2 would still be a portable console, it either takes concessions (resolution) or a lot of work to port it. Thats simply the reality, and nintendo(nvidia) cant create a magic chip that circumvents that.

Yeah, it's extremely capable. Almost as good as drawing the pixels myself.

I like your pixels more, have a more personal touch

Not really. You have to do all that regardless, you need a complete rendering pipeline that generates a totally complete image before DLSS can function.

True(except couldnt you do that for parts, and in paralell while those get dlss'ed render something else in lower resolution. In parallel and compose the 2 images? (Just spitballing).
Like rendering the background, upscaling, and render the character models (assuming that they are not screen filling and with that way less pixels to render, filling the rest with transparency)

DLSS 2 has one advantage over native rendering, which is subpixel stability - the fizzing that can happen when lots of highly detailed objects, like foliage, move on screen. But that's not a clear win when the rest of image quality is coming down.

...honestly, thats one of my BIGGEST gripes. Those annoy and distract me way more then less details. For me it would lead to a more pleasant image. But AA can do that to, its just rarely seen on the switch.

DLSS 2 still runs on the GPU as a whole, it's just that it's core uses special instructions the tensor cores make fast. You can overlay CPU and GPU work like this - in fact, that's how DLSS 3 works - but I'm not sure you can overlay Tensor Core and GPU work like this in any way that is a significant win.

Oh, you cant? Ok, thats one thing i did not know. Thanks for that info

 

[Truno]

I have a feeling that Nvidia itself wants the Switch 2 to be a powerful console

NVIDIA wants Nintendo's money

 

[ILikeFeet]

I have a feeling that Nvidia itself wants the Switch 2 to be a powerful console

they can always give nintendo discounts for that

 

[Hermii]

they can always give nintendo discounts for that

They never would though. They are a business, not fans.

 

[Paul_Subsonic]

Does anyone know how to denoise and sharpen that ?

 

[Kevin]

Why does the fact that it uses a fanmade cover art say the date is wrong?

I mean, it's probably wrong of course but I don't see this meaning so.

Oh I was just sharing that they used a fan made box art. It's not 100% evidence that the date is fake but does make one a bit more skeptical since the box art is wrong. Probably just a placeholder date.

 

[Gerald]

NVIDIA wants Nintendo's money

Ik, but Tegra X1 in switch 1 must give them very big profit

 

[Deleted member 887]

I have a feeling that Nvidia itself wants the Switch 2 to be a powerful console

NVIDIA wants Nintendo's money

they can always give nintendo discounts for that

They never would though. They are a business, not fans.

Nvidia wants everyone's money. And they don't just want Nintendo's money today they want it forever.

Nvidia wants every data center in the world to be full of Nvidia machine learning hardware. They've done it partially by unifying their data center and consumer lines, battle testing their designs at massive scale from both ends, and delivering generational leaps in the data center at a consumer pace. But to keep that going, they need to keep pushing the hardware further, and justifying it to their consumer customers.

Nvidia wants every game possible to ship with an excellent DLSS implementation and Ray Tracing features, in order to maximize the appeal of their hardware, and continue to justify their investment in ML and RT.

Meanwhile, AMD's grip on consoles means that, despite having a minority position in graphics cards, lots of games ship with engines tuned to AMD's hardware - see Death Stranding for the most extreme example.

On the other hand, Nvidia wants every device with a camera to ship with one of their SOCs. Your vacuum cleaner robot, your car, every robot arm in an assembly line, every security camera in a grocery store, powered by Nvidia computer vision hardware and an onboard CPU, sipping power and running single software defined architecture that marries easy-to-develop-for and vendor-lock-in for decades of revenue.

Getting games designed to run on low power hardware to ship with RT features? A win.

Making every multiplat game support DLSS? A win.

Building up their SOCs to be a rival to AMD's APUs? A win.

Delivering a generational leap to Nintendo, along with a feature set that is hard to mimic on a rival's hardware? A win.

I'm sure that Nvidia is going to make money off of this deal, and a lot of it. But they have every incentive to drive their initial margins down to nothing

 

[Gerald]

Nvidia wants everyone's money. And they don't just want Nintendo's money today they want it forever.

Nvidia wants every data center in the world to be full of Nvidia machine learning hardware. They've done it partially by unifying their data center and consumer lines, battle testing their designs at massive scale from both ends, and delivering generational leaps in the data center at a consumer pace. But to keep that going, they need to keep pushing the hardware further, and justifying it to their consumer customers.

Nvidia wants every game possible to ship with an excellent DLSS implementation and Ray Tracing features, in order to maximize the appeal of their hardware, and continue to justify their investment in ML and RT.

Meanwhile, AMD's grip on consoles means that, despite having a minority position in graphics cards, lots of games ship with engines tuned to AMD's hardware - see Death Stranding for the most extreme example.

On the other hand, Nvidia wants every device with a camera to ship with one of their SOCs. Your vacuum cleaner robot, your car, every robot arm in an assembly line, every security camera in a grocery store, powered by Nvidia computer vision hardware and an onboard CPU, sipping power and running single software defined architecture that marries easy-to-develop-for and vendor-lock-in for decades of revenue.

Getting games designed to run on low power hardware to ship with RT features? A win.

Making every multiplat game support DLSS? A win.

Building up their SOCs to be a rival to AMD's APUs? A win.

Delivering a generational leap to Nintendo, along with a feature set that is hard to mimic on a rival's hardware? A win.

I'm sure that Nvidia is going to make money off of this deal, and a lot of it. But they have every incentive to drive their initial margins down to nothing

COD also work better on AMD GPUs, many games and engine, thanks to consoles

 

[ShadowFox08]

16RAM will be great, imo it will be 2GHz

I'm not betting on either. For one, For the RAm situation... 16GB might just be overkill for Switch 2/Drake. 12 is actually the sweet spot that a lot of us are hoping for, and it will be more likely than 8GB and 16 GB, especially If it's a 2024 release (assuming soc isn't taped out)... If they go with an OS ther only uses 1-2GB, they will have 10GB left for games, which is plenty enough for it's power profile. it's gonna be more efficient than the Series S.
Second, 16GB is also more costly than 12GB in price and power. I'm all up for future proofing this thing, but I really don't think it needs it. SD handheld kinda needs it cause it's a beefy OS.

What's more important is the bandwidth for Drake. 102 GB/s should (imo) be enough for 1080p PS4 ports without DLSS, based on what I've seen for multiplat ports on switch that had 1/4 of that, and Nvidia being really good with their compression (as well as Drake supposedly having decompression hardware like current gen) + increased cache. The more the better though. 133 GB/s would be the best case scenario. At least most 4k native switch games should ne possible with 102 GB/s without DLSS,

For the CPU part..
While 2Ghz CPU would be absolutely amazing, I'm not expecting it. Could it be done on 4nm TSMC with a 2.5-3 tflops GPU? Maybe. I don't know. It's a huge power savings jump from 8nm samsung. But Nintendo will prioritize balancing life battery first.

CPU and Bandwidth are what we need the most for getting current gen ports, and running at acceptable levels. It's gonna be the choke points/bottlenecks again, just like Switch. Having a 1.7-2Ghz CPU will be better in the long run for Switch 2 than a 3-4 tflops GPU. Like I would settle for a 2.5 Tflops GPU for the highest CPU. At least GPU scales nicely, and we have DLSS. Though it will be interesting to see how AMD's version of upscaling will evolve and be utilized against current gen consoles to match it.

Single core performance is roughly the same, but Zen 2 pulls ahead as the clock goes higher - sustained server benchmarks tend to show advantages for Zen at high clockspeeds. The fact that Drake will almost definitely be clocked way below the other consoles exacerbates that gap.

AMD CPUs are great at multithreading. AMD can shove huge numbers of cores onto a socket, and each of those cores supports two threads. ARM CPUs only support one thread per core.

In the Zen 2 case, threads on the same core essentially can share data for free, but on different cores they have to copy data. That's the ARM advantage - the whole cluster can share data for free. In this exact situation, where it's 8 Zen 2 cores versus a single cluster of 8 ARM CPUs, that's advantage ARM on multithreading.

Desktops and Servers often have lots and lots of threads devoted to wildly different tasks on the OS side, touching different pieces of data. Video games love to have lots of threads all try to access a single, global piece of data about the game world. That's why Drake is able to make more of the multithreading advantage than you might think on paper.

I do remember (was it you) who posted benchmarks. Single core wasn't that far off, but I wasn't sure if it was per GHz matching A78 with Zen 2.. yeah The lower speed for the A78c CPUs is the expected drawback like you said.. Kinda surprised about the multi core performance though.

Btw, What power advantages does A78c have over a78e CPUs anyway?

They never would though. They are a business, not fans.

Nintendo is definitely getting some kinda discount. They already have a good relationship with them, and they're ordering millions. But that's kinda standard business though. But we're already taking that into account and expecting a $400 price at least anyway due to inflation.

I wonder how much a dev kit identical to Drake would cost. A lot more of course.. weren't the AGX models going for like a grand or something?

 

[NineTailSage]

I suspect we'll see ultra performance mode as well for 4K output. The struggle is that DLSS's performance curve scales (mostly) with output resolution, not scaling factor. So 4K output costs the same (almost) regardless of what base resolution you start with.



Absolutely true. I took the ~5ms numbers from the GDC data a few years back about post-processing effects and frame times. But yes, it is variable. There is that constant tug between "explaining why these numbers are eyebrow raising" and "laying down all the nuance."

I think @Paul_Subsonic's analysis is solid, but I can't imagine enabling hardware accelerated DLSS on hardware where it is effectively unusable. So I'm pointing in the direction of "very small number of usable benchmarks" and shrugging my shoulders until there is more data to go on.

This was taken from Digital Foundry and their Death Stranding analysis.(I'm actually surprised that Nvidia doesn't have any updated charts about DLSS render time)

"On an RTX 2080 Ti at 4K, DLSS completes in around 1.5ms - meaning it's faster than checkerboarding on PS4 Pro. However, the hardware comparison is obviously lopsided. The weakest capable GPU is the RTX 2060 (still significantly more powerful than the Pro) and DLSS has an overhead in excess of 2.5ms on this card. That's heavy, especially if you're targeting 60fps where the entire frame rendering budget is just 16.7ms. However, the core advantage is that the base resolution is so much lower. DLSS in Death Stranding comes in two flavours: the performance mode achieves 4K quality from just a 1080p internal resolution. Meanwhile, the quality mode delivers better-than-native results from a 1440p base image. In both cases, that's much lower than PS4 Pro's 1920x2160 core resolution. By running everything else in the GPU pipeline at much lower resolutions, the cost of processing DLSS is more than offset - to the point where mildly overclocking the RTX 2060 allows Death Stranding to deliver 4K gaming at 60fps."


Also in this Wccftech on DLSS they include all of the charts and the numbers that are factoring in seem to be using 540p as the base resolution. Hence why they show rendering times for 1080p, 1440p and 4k.

NVIDIA DLSS 2.0 Behind the Scenes - How the Magic Happens

NVIDIA published a GTC 2020 talk where Senior Research Scientist Edwin Liu revealed how DLSS 2.0 can achieve such great results.

wccftech.com

 

[Paul_Subsonic]

Forget about my post asking to sharpen some blurry stuff, looks like the chinese version is freely accessible.
If I can translate it, we can have access to all the info we need on DLSS.

Edit : well the post above has the full res screenshots of that GTC Talk, so no need to sharpen blurry stuff nor translate chinese.

 

[Paul_Subsonic]

Stuff that may be interesting :
1) In the Chinese version of the GTC, they're talking about DLSS2.1 but the numbers in the graph are the same. No performance difference there.
2) We can use techpowerup DLSS reviews to measure performance difference between DLSS versions.
3) We may actually not need to do that because with the techpowerup DLSS reviews, we can compare 1080P native vs 4K DLSS, and techpowerup have the courtesy of showing the clock speeds of the GPU, making the data extremely precise.

 

[Dekuman]

Nvidia wants everyone's money. And they don't just want Nintendo's money today they want it forever.

Nvidia wants every data center in the world to be full of Nvidia machine learning hardware. They've done it partially by unifying their data center and consumer lines, battle testing their designs at massive scale from both ends, and delivering generational leaps in the data center at a consumer pace. But to keep that going, they need to keep pushing the hardware further, and justifying it to their consumer customers.

Nvidia wants every game possible to ship with an excellent DLSS implementation and Ray Tracing features, in order to maximize the appeal of their hardware, and continue to justify their investment in ML and RT.

Meanwhile, AMD's grip on consoles means that, despite having a minority position in graphics cards, lots of games ship with engines tuned to AMD's hardware - see Death Stranding for the most extreme example.

On the other hand, Nvidia wants every device with a camera to ship with one of their SOCs. Your vacuum cleaner robot, your car, every robot arm in an assembly line, every security camera in a grocery store, powered by Nvidia computer vision hardware and an onboard CPU, sipping power and running single software defined architecture that marries easy-to-develop-for and vendor-lock-in for decades of revenue.

Getting games designed to run on low power hardware to ship with RT features? A win.

Making every multiplat game support DLSS? A win.

Building up their SOCs to be a rival to AMD's APUs? A win.

Delivering a generational leap to Nintendo, along with a feature set that is hard to mimic on a rival's hardware? A win.

I'm sure that Nvidia is going to make money off of this deal, and a lot of it. But they have every incentive to drive their initial margins down to nothing

Yeah nvidia and Nintendo are a good fit for this reason. Strangely because AMD's strong position in the console space. This however means Nintendo gets nvidia's undivided attention. It's why I don't really see them splitting up even though I always entertain that every six months when they don't announce the succ, or most recently, when reports of the revision being scrapped surfaced.

 

[Gerald]

I'm not betting on either. For one, For the RAm situation... 16GB might just be overkill for Switch 2/Drake. 12 is actually the sweet spot that a lot of us are hoping for, and it will be more likely than 8GB and 16 GB, especially If it's a 2024 release (assuming soc isn't taped out)... If they go with an OS ther only uses 1-2GB, they will have 10GB left for games, which is plenty enough for it's power profile. it's gonna be more efficient than the Series S.
Second, 16GB is also more costly than 12GB in price and power. I'm all up for future proofing this thing, but I really don't think it needs it. SD handheld kinda needs it cause it's a beefy OS.

What's more important is the bandwidth for Drake. 102 GB/s should (imo) be enough for 1080p PS4 ports without DLSS, based on what I've seen for multiplat ports on switch that had 1/4 of that, and Nvidia being really good with their compression (as well as Drake supposedly having decompression hardware like current gen) + increased cache. The more the better though. 133 GB/s would be the best case scenario. At least most 4k native switch games should ne possible with 102 GB/s without DLSS,

For the CPU part..
While 2Ghz CPU would be absolutely amazing, I'm not expecting it. Could it be done on 4nm TSMC with a 2.5-3 tflops GPU? Maybe. I don't know. It's a huge power savings jump from 8nm samsung. But Nintendo will prioritize balancing life battery first.

CPU and Bandwidth are what we need the most for getting current gen ports, and running at acceptable levels. It's gonna be the choke points/bottlenecks again, just like Switch. Having a 1.7-2Ghz CPU will be better in the long run for Switch 2 than a 3-4 tflops GPU. Like I would settle for a 2.5 Tflops GPU for the highest CPU. At least GPU scales nicely, and we have DLSS. Though it will be interesting to see how AMD's version of upscaling will evolve and be utilized against current gen consoles to match it.


I do remember (was it you) who posted benchmarks. Single core wasn't that far off, but I wasn't sure if it was per GHz matching A78 with Zen 2.. yeah The lower speed for the A78c CPUs is the expected drawback like you said.. Kinda surprised about the multi core performance though.

Btw, What power advantages does A78c have over a78e CPUs anyway?


Nintendo is definitely getting some kinda discount. They already have a good relationship with them, and they're ordering millions. But that's kinda standard business though. But we're already taking that into account and expecting a $400 price at least anyway due to inflation.

I wonder how much a dev kit identical to Drake would cost. A lot more of course.. weren't the AGX models going for like a grand or something?

TSMC 4N will be awesome for switch 2, i think for sure switch 2 will be have bigger battery, and its important to have ports from PS5 and Xbox Series X|S, teraflops idk but 4teraflops will be the best scenario and enough

 

[Hermii]

Nintendo is definitely getting some kinda discount. They already have a good relationship with them, and they're ordering millions. But that's kinda standard business though. But we're already taking that into account and expecting a $400 price at least anyway due to inflation.

I wonder how much a dev kit identical to Drake would cost. A lot more of course.. weren't the AGX models going for like a grand or something?

Yea, that's business. What I meant though, is they wouldn't give them a discount just because they want it to be powerful or whatever.

 

[Dekuman]

Nvidia's value to Nintendo would be on the NVN APIs and the SoC tech as a package. I don't think you can divorce the two. It makes it harder for Nintendo to go back to AMD or another vendor.

 

[Dakhil]

Btw, What power advantages does A78c have over a78e CPUs anyway?

I don't know if the Cortex-A78C has any power advantage(s) over the Cortex-A78AE.
But I think the Cortex-A78C does offer a latency advantage over the Cortex-A78AE since the Cortex-A78C can have 8 CPU cores per cluster whereas the Cortex-A78AE needs two clusters to have 8 CPU cores in total since the Cortex-A78AE can only have 4 CPU cores per cluster. Therefore, there's less hardware components needed to communicate with if the Cortex-A78C is used instead of the Cortex-A78AE.

 

[Hermii]

Nvidia's value to Nintendo would be on the NVN APIs and the SoC tech as a package. I don't think you can divorce the two. It makes it harder for Nintendo to go back to AMD or another vendor.

i agre. Sure Qualcomm (for example) could provide great hardware, but they are far behind on the software bit.

Nintendo was never know for having great tools prior to the switch.

 

[Dekuman]

i agre. Sure Qualcomm (for example) could provide great hardware, but they are far behind on the software bit.

Nintendo was never know for having great tools prior to the switch.

yeah. i had written something about their experience with AMD vs nvidia being night and day before deleting it, realizing, the frakenstein Wii U McM probably had more to do with the inherited partnerships between AMD/IBM than AMD's fault. I'm sure AMD could have offered up their own CPU/GPU combo for the Wii U that was cheaper and more performant than what we got, but one thing AMD didn't offer was good tools.