Is there a limit to how much you can shrink a processor before it breaks compatibility? If we were to assume that Nintendo must stick with the Maxwell architecture for backwards compatibility, how small of a process can they shrink that architecture before it starts to behave differently? Maxwell and Pascal are pretty much the same, but the smallest process I have seen a Pascal based GPU manufactured on is 14nm. With the move to something like 7nm or even 5nm, the GPU core count could be tripled and still fit within the power/thermal limitations of the Switch form factor.
The next Nintendo hardware? [2021-09]
- Thread starter wco81
- Start date
Not really, no.
In most cases, over the time span required to move several process nodes and given the much larger transistor budgets it has historically enabled, architecture typically evolves a bit, but that doesn’t mean that it has to. And even if they did Maxwell, that doesn’t mean they couldn’t beef it up in compatible ways from the rather straightforward (larger caches) to the less so, but still modest in engineering cost compared to rolling out a 5nm SoC in the first place.
Honestly, if I were Nintendo, and it was necessary to achieve seamless backwards compatibility, I would do something like that. Tech enthusiasts would be horrified of course, but it would probably work out nicely for Nintendo and consumers both.
Or they could simply leverage the fact that even a remotely modern SoC would be much more powerful and simply emulate the current Nintendo Switch if people really are screaming for backwards compatibility. It's not something Nintendo have cared about historically and I fail to see why it suddenly would be important now.
I don't personally care and think they should just start with a clean slate and perhaps port their own 1st party games (which is the reason to own an Nintendo console to begin with) if need be. Or release with a strong game library of 1st party games and get the next generation of handheld gaming going.
I don't personally care and think they should just start with a clean slate and perhaps port their own 1st party games (which is the reason to own an Nintendo console to begin with) if need be. Or release with a strong game library of 1st party games and get the next generation of handheld gaming going.
A new SoC on 5nm would be a lot more powerful regardless of underlying graphics architecture. It would fit roughly four times the gates of the TX1 in a chip roughly 60-70mm2 in size.
The reason for making an extension of Maxwell would be backwards compatibility and protect the investment (financially and emotionally) of their customers.
If that is a priority and if that requires the next SoC being based on Maxwell, then it makes sense, (and would likely do the job just fine). But the reasoning hinges on conditions we don't know are fulfilled. Goodtwin simply asked if it was possible extend Maxwell to 5nm, to which I responded "Yeah, sure." It was not a prediction.
Johnny Awesome
Veteran
From a business standpoint, Nintendo should do what they always do - break backwards compatibility and make everyone pay for the same games all over again on a new underpowered platform. 
Nintendo has actually been a big on backwards compatibility. The Gameboy Advance could play Gameboy games, the DS could play Gameboy Advance games, the 3DS could play could play DS games, the Wii could play Gamecube games and the Wii U could play Wii games. This was all done in hardware and not through emulation, correct me if I am wrong. Nintendo has more often than not extended backwards compatibility to the prior generation.
This is why I cant shake the idea that its not only possible Nintendo will do this, but actually very plausible. Especially if the newer process allows for 3-4x the number of GPU cores. There is no doubt the newer Ampere will outperform the Maxwell flop for flop, but to what extent? Most of the buzz in the gaming world surround ray tracing and would require hardware that they simply wont be squeezing into an SOC that needs to operate on 6 watts in portable mode. Im willing to bet having flawless backwards compatibility will be a requirement from Nintendo and this seems like a simple way to go. What is Nintendo's target performance envelope for Switch 2? Its not going to be eyeing up the PS5/X, there is no mobile hardware that can even get in the same zip code as those consoles. So its really about offering a meaningful upgrade over the current Switch. PS4 level visuals running at 720p portable mode on an OLED screen would be a meaningful upgrade. Nintendo already patented some sort of bespoke DLSS in 2020, my assumption is that is how they plan to improve Docked presentation when being played on 4K screens.
So, in summary, they've done both. And both approaches have merits and drawbacks that need to be balanced, depending on market conditions, design goals, available technology and cost.
The problem with speculating around Nintendos next step is that they have so many reasonable options. For instance - backwards compatibility or not. And if so, via some (imperfect) emulation, underlying hardware architecture or a mix?
What process, what supplier, what memory subsystem solution, what cost constraints, what power draw...
That said, I tend to agree with your assessment. If Nintendo wants good backwards compatibility, and they move to TSMC 5nm (or something close to it), then going with an (improved) Maxwell architecture would probably be a good solution overall, if moving to the latest Nvidia GPU architecture would create problems. Did I sprinkle enough conditionals in there?
Fundamentally, I expect them to be memory subsystem limited to a large degree. Only Apple has a greater than 64-bit wide memory subsystem on a mobile device, and only on their iPad Pros, so my guess is that Nintendo will stick to a 64-bit LPDDR5(x?) memory subsystem. Which in turn implies that they will try/have to alleviate that bottleneck with greatly improved cacheing. Of course cache carries its own costs in die area and power draw so there is a balance to be struck, but even with sharp cost constraints 5nm still allows for a huge expansion over the TX1. (Apple A15 die shot and analysis. Note L2 and LLC capacities and die areas.)Presumably, this scenario would mean no RT and no DLSS, but I can't see that as significant for their product or target demographic. Why would Nintendo customers care exactly what computational scheme is used to calculate their ambient occlusion? And DLSS is arguably not the best fit for a small mobile SoC working over an already uncomfortably thin memory pipe, never mind the installed base of software and existing projects - remember this whole thought experiment hinged on backwards compatibility, so the upscaling method should ideally be functional across the board. The saved die area can be used to either simply reduce cost or to beef up other areas of the SoC.
So - under a given set of assumed constraints and parameters, I'd say that this would make sense for Nintendo and its customers, and preserve investments in software for all involved while providing a reasonable generational leap in performance. But if I changed even one of those conditions, in this case most glaringly the need to stay with a Maxwell(-derived) architecture for backwards compatibility, then the picture changes. Or if they go with, say, Samsung 8nm process. Or...
It's just another thought experiment.
The Tegra X1 Parker already moved to a 128 bit memory bus, so I think its feasible they would implement that with the new SOC. Switch isnt quite as hamstrung by thermals and battery life as a smart phone is, so running a bit hotter than you might spec for a true mobile SOC is acceptable. I believe I remember seeing some benchmarks on the Tegra Parker chip a few years back, and it was quite staggering how much the increased bandwidth improvement the performance compared to the standard X1. Combine a 128bit bus with LPDDR5 memory and we start to get into the Xbox One level of bandwidth. A mobile processor will never be matching what the space heater consoles can manage, but if Switch 2 multiplies the GPU cores by 3-4x without bringing a substantial boost to memory bandwidth, its not going to see the full benefits of the additional cores. Especially if they move to 8 CPU cores where there will be even more contention for that limited bandwidth.
I think at this point, when people assume Switch 2 will use DLSS, they really just mean some sort of image reconstruction. DLSS specifically leverages there tensor cores included in Nvidia graphics cards to do the work, and they do it very well. However, just like AMD created Fidelity FX, I'm certain Nvidia can implement something similar that plays nicely with the new SOC. Its pretty much the only way Nintendo will be able to offer any sort of 4K docked experience.
To be clear, I am not advocating that this is what I want. As a gaming enthusiast, I would rather see Nvidia put together a cutting edge SOC that implements their latest technologies and still maintains BC with the original Switch software. I went 100% digital with my games on Switch and have a library of over 100 games. It would be a kick in the teeth to not be able to play those on the next Switch. Having said that, after months and months of debate over the Wii U GPU, once we finally came to a consensus, Nintendo will make hardware choices that seem asinine to many tech enthusiast. Some of Nintendo's greatest successes were powered by hardware that was far from cutting edge. Regardless if its Maxwell or Nvidias latest architecture, the Switch 2 will sell based on its software and its ability to play on portably and docked.
Indeed. And it's not that a 128-bit interface to LPDDR5 is totally impossible, as demonstrated by the Steam Deck. But it represents a large step in all aspects - it increases power draw, increases die area directly to drive, but even more so since you would tend to scale at least the GPU resources to take advantage of it. It would mean not only a factor of two in bandwidth, it would pretty much imply a factor of two across the board. Which, if you are exclusively performance oriented would be great, but it would clearly raise questions regarding form factors (Switch Lite and Switch have roughly cell phone capacity batteries - which helps with cost, size and weight), and of course total cost of the device, and Nintendos options when it comes to produce wallet friendly models and increase installed base.
We'll see, but I expect Nintendo to be conservative.
Well, 4k TV sets are, depending on set size and viewing distance, close to or beyond the ability of the eye to resolve pixels. Under those conditions, I'd contend that absence of artefacts are more important than pixel level quality.
We have come a long way in upscaling from nearest neighbour and bilinear. The web is full of comparison images for photographers. And these methods use no temporal data at all. They would all be applicable as a final step before output, regardless of source. But they are arguably overkill, under the conditions we are discussing even something like bicubic+mild sharpening looks very good. Could you see differences when blowing up the image 400% and looking at stills? Sure. Does it matter when you're playing a game and the output resolution is close to or even beyond your ability to resolve? IMHO, no.
Not that my opinion is worth more than you paid for it, but I agree that not having backwards compatibility would be perceived as disappointing by their current (huge) user base, and would expose Nintendo to greater financial risk. It would be easier to justify if the new console was something completely new such as standalone VR-glasses. But if it is pretty much an improved Switch, then backwards compatibility would be expected.
Actually no. The M1 uses LPDDR4x. Only the M1Pro and Max uses LPDDR5.
Thanks for the correction. The point still stands though, that there is nothing standing in the way for making SoCs that use LPDDR5 with a wider memory bus.
Nintendo broke backwards compatibility with a console almost no one bought -- I had a wiiu, but how many of you did? Typically they provide bc with their followups to popular devices, so I'd expect it to be a priority for the switch successor.
Other than cost and power.
I think Steam Deck is the better example, a much closer related device, and at a reasonably close price point. So yes, it's possible. Whether it's likely is another question. Much as I like the Steam Deck for what it is, I don't think either Nintendo or its customers would like a Switch 2 modelled after it. They'd prefer smaller, lighter, cheaper, better battery life. Going all out on SoC performance fits that list like a foot in a glove.
Gotta admit, it would be awesome if Nintendo was able to just build Switch 2 around an Apple A14 or A15.
Are you not a fan of image reconstruction then? Have we seen any examples where developers have used some of the alternatives you described?
4K would largely be irrelevant if the average size for a TV hadn't increased so drastically. I currently game on a 42" LG 1080P LCD TV sitting roughly 9ft from the TV. A 4K display at the same size and viewing distance will look nearly the same. With that said, many consumers are choosing to purchase the largest set that fits within their budget and even 70+ inch displays are very affordable these days, and end up in rooms that have the user sitting less than 10ft away, a distance where 4K is a meaningful upgrade.
Apple products are priced much higher than Nintendo will sell the Switch 2 for. Apple may very well create some SOC's that are using large memory buses, but are any of them used in a sub $500 device? With that said, the Tegra Parker chip moved to a 128 bit bus without drastically altering the size of the SOC. At some point with the Switch 2 being a gaming only device, perhaps sacrificing some GPU cores in favor of more memory bandwidth would offer superior results. No point in becoming completely unbalanced. If Nvidia decides they can tackle the problem another way like Entropy described with larger ALU cache or perhaps something similar to AMD's infinity cache, then perhaps sticking with a 64 bit memory bus will be sufficient.
Is there a hardware arch that'll scale down to a Nintendo level portable and up to decent home console? I think Nintendo will prioritise a portable SOC architecture for the handheld, so you'd need something that'll scale up from that. But even then, Nintendo don't seem happy abstracting their hardware yet, so I think you'd need scalability to handled mostly in hardware with Nintendo writing their games for portable and these automagically being mapped onto a different SOC with more execution units.
The thing is that there are a number of variables, as usual it is a complex question that tends to get simplified (to something like "best possible image quality when upscaled a factor of two").
Does it require extra hardware? How much? Does the method put extra pressure on other parts of the system, like the memory channel? Is it applicable to all new games? Is it applicable to already existing games? What are the expected output devices and viewing distances?
In my book, upscaling in a games context is potentially quite useful, particularly as the resolution of the output devices are getting to the point where the native resolution is close to or even beyond our ability to resolve. Our perception of detail is dependent of brain processing and also gameplay is dynamic, so evaluation via full image error metrics of still images as is common in for instance research focussing on photographic images is of limited use (as is subjective evaluation of enlarged images).
My subjective opinion is that in gaming, today, we don't really have reason to aim for maximum quality in upscaling from lower resolution. Rather, we should be looking at maximising utility at minimum cost in processing (silicon/frame time/latency). But the discussion on all gaming tech oriented fora has been skewed by NVidias desire to sell DLSS as a unique feature of their GPUs. I think that is a bit unfortunate. In my personal opinion, it would be also be nice if the method was universally applicable, that is, even to old titles, and that the compromises implicit with that goal are worth it since maximum quality is irrelevant for the application.
In the context of speculation about Nintendos next console hardware, assuming they need to stay within existing cost and power constraints, I could see them going with something simple and efficient if that matches their goals. But just what those goals would be is anybodies guess. I would be somewhat disappointed if it only applied to a subset of new titles and all other software was left to whatever the TV-set manufacturer decided to use for their game mode setting (as is the case on 4k TVs at the moment).
I do not see them doing that. Nintendo does not like to take losses on hardware and thus will not be able to compete with the performance that Sony and Microsoft offer at a given price point. Nintendo has done well in the portable market for decades but home console has been trickier for them. Switch has allowed Nintendo to offer their entire offering of new software on a single platform, and the subset of the audience that are interested in Nintendo games but only if they are on a powerful home console is minimal. The hybrid setup has worked for Nintendo and I assume they will give it another go for at least one more generation.
I agree that whatever they choose will be simple and probably light weight in terms of performance cost. I am thinking it will be applied through system settings. Switch 2 will likely still be limited to 1080p rendering for docked gameplay and with each passing day there are more and more consumers replacing their old 1080p TV's with new 4K models. Having some sort of solution to improve presentation for a 1080p rendering resolution (or even lower) on a large 4K set is the goal. DLSS requires tensor cores, so that may be impractical for a mobile SOC, we don't really know yet. Regardless, the objective would be to get superior results compared to simply sending the raw native resolution to the TV and letting that scale the image.
Similar threads
