I dunno , I think it may be a package deal , ray tracing and dlss will be in the same core. I could be wrong.
Switch 2 Speculation
- Thread starter Goodtwin
- Start date
I dunno , I think it may be a package deal , ray tracing and dlss will be in the same core. I could be wrong.
D
Deleted member 13524
Guest
Not in the case of a Volta GPU. And we don't know if nvidia couldn't implememt DLSS in the lower end Turing chips that apparently have some form of cut-down tensor cores for FP16 throughput.
How will having features such as RT or DLSS automatically help developers improve their art pipelines ?
Did RT hardware in the Xbox Series S help out developers at all in the end when none of them are using it over there currently ? If a system like that featuring 4+ TFlops of power couldn't even get ray traced graphical effects featured into it's games then what hope could an upcoming low power SoC that won't even come close possibly have to change that outcome ? Retaining developers with a hardware feature that won't even be used in games might as well be a lost cause ...
As for DLSS, this is not a helpful feature for artist pipelines and is actually harmful to them because it places technical restrictions to art design or other rich/complex features to the pipelines. DLSS works best when it can work with a relatively high internal resolution with high quality settings and simpler art pipelines. You lose visual consistency with lower settings and lower internal resolutions along more complex art pipelines so how will DLSS work out on an SoC that will likely be at the lower end of those parameters ? DLSS could very well be a better fit for the simpler art pipelines that Nintendo uses for their games which will minimize visual inconsistencies but what are the chances of other developers compromising their own more complex art pipelines ?
I think a lot of the views in here behind features like RT or DLSS are misguided with respect to how they impact a developer's willingness and how they affect the ability of their art pipelines to produce content on other platforms ...
RT is used with Series S. IDK what you're talking about. Watch Dogs uses it, for example and it looks beautiful for that price. The upcoming next gen version of Metro Exodus will run exclusively on RT capable hardware, and it runs with RT on the Series S at around 1080p60. This is one game that would have not been possible without RT acceleration, so a Switch Pro port of that game only could be doable with RT acceleration. Meaning RT acceleration is crucial for next gen support. Keep in mind Switch Pro would have DLSS which would overcome the huge power differences between the Series S and the Switch Pro just nicely, plus Switch Pro's RT cores are faster than AMD's Ray accelerators, further closing the gap. It is totally in the realm of possibility.
Also I don't know what you are talking about DLSS. We have seen a magnitude of titles, especially UE4 which varying asset quality and art styles, DLSS works just as good as with more demanding titles. Same goes for resolution, it does a great job at lower resolutions as well, you can watch DF's video on that matter. And there are many videos on YouTube where DLSS is asked to reconstruct a crazy low res like 360p.
Thinking about it, it makes perfect sense for Nvidia to deliver the RTX ecosystem into the console space by using the Switch Pro. I assume they are going to subventioning the hell out of the SoC for Nintendo for that purpose.
Also I don't know what you are talking about DLSS. We have seen a magnitude of titles, especially UE4 which varying asset quality and art styles, DLSS works just as good as with more demanding titles. Same goes for resolution, it does a great job at lower resolutions as well, you can watch DF's video on that matter. And there are many videos on YouTube where DLSS is asked to reconstruct a crazy low res like 360p.
Thinking about it, it makes perfect sense for Nvidia to deliver the RTX ecosystem into the console space by using the Switch Pro. I assume they are going to subventioning the hell out of the SoC for Nintendo for that purpose.
Last edited:
Look. I would rather have the real resolution vs upscaling any day of the week. However right now the switch 1 is a 720p at best machine portable and a 900p at best machine docked while firmly behind the xbox one in terms of performance .
If Nintendo is going to make another console I don't see them putting in 4k capable hardware (and I mean this in terms of playing games like current gen watch dogs or AC or COD). What to me seems most likely is them targeting 720p in handheld or maybe stretching to 900p in handheld. That means most likely 1080p in docked mode is the upper limit of what we would see
So the question is do you want a 720p to 1080p out put on a 4k tv and leave it up to the scaler on the tv or would you rather use something like DLSS ?
Now like I said I would love to have a 4k portable system. It just wont happen. But I would settle for a 720-1080p source using DLSS to 4k vs just a 720p or 1080p source on a 4k tv.
Yes this is important to think about. A lot of the switch success lies in the fact that it was a huge port machine esp for indies and lower budget games. It may not seem like an issue now in 2021 but both xbox and playstation as well as pc have RT and as we move forward in the time line more and more gamers will have RT capable hardware be it through graphics card upgrades or new system purchases. More games will just use RT or more be designed with only RT in mind.
of course this is nintendo and they do crazy things all the time and currently there is no other big player in the switch like segment. Yes you have some smaller companys making thins like the Aya neo and there is a rumor of Qualcom putting out a switch like unit early next year. However none of them really have their own exclusive software nor the scale of a new switch.
I think having just a switch + will put nintendo in a difficult place in the next few years
I mean does it matter ? Nintendo can just do what MS is doing. For 2 years or so First party games launch on both systems and 3rd parties can pick what they want to do and we see a gradual slow down of the switch.
Frenetic Pony
Veteran
DLSS isn't magic, the new gen non portables get good upscaling too. Looking at UE4's new upscaler, frankly it has less errors than DLSS for a lower performance cost. I'm sure AMD's thing will be fine as well.
But the reality of ports is much worse, or much better, than the GPU regardless. The real hurdle that a Switch 2 simply won't overcome is CPU porting. Zen 2 cores are modern, high performance, and on all 3 consoles high clocked. What's more is the consoles have dedicated hardware for sound and decompression on top of that. A lot of "big" triple A titles are either going to use that power, be first party only titles (witness Square Enix seemingly being bought in practical terms by Sony/MS buying Bethesda) or...
Or this is the other major thing, the games won't care about the CPU, and probably not the GPU that much either. A lot of the time the best seller list on Steam is stuff that might be ported to Switch 1 anyway. Not that a Switch 2 is a bad idea, but that "4k" is both a hilarious claim (what are you supposed to do, upscale 720p at best?) and that it's also an unneeded claim. Nintendo would do better to not concentrate on such things, even as Nvidia pushes them to. I mean, I know that's what this forum is for. And I love technical stuff. But as stated, it's a bad claim, and as far as business, and actual games are concerned. Well the last game I played for hours on end was also a hit best seller. It was Loop Hero, a pixel art mashup that'd probably be portable to the DS, or lower.
Frenetic Pony
Veteran
All that being said, I am interested in seeing Ada, if that's really what's in it. It would explain the higher end performance to energy usage claims. RT I can't see happening, everyone on here seems to forget how ridiculously expensive it is in comparison to the rest of the pipeline. As in "Nvidia's top end cards can run last gen games with single raytraced effect on top!" level of bad. Thus a console barely more powerful than last gen consoles to begin with running RT doesn't seem likely, unless Nvidia has somehow vastly increased RT efficiency.
But, Nintendo versus Microsoft? (Cloud streaming games doesn't seem to be going great for anyone). Later last gen ports to mobile, ooh. And it definitely would open up more ports besides. Valheim doesn't interest me, but I'm sure plenty of people would love to see it in portable. How about Baldur's Gate 3? Etc. A lot of cool stuff definitely could come up.
But, Nintendo versus Microsoft? (Cloud streaming games doesn't seem to be going great for anyone). Later last gen ports to mobile, ooh. And it definitely would open up more ports besides. Valheim doesn't interest me, but I'm sure plenty of people would love to see it in portable. How about Baldur's Gate 3? Etc. A lot of cool stuff definitely could come up.
How can we be certain that the Series S will be running at those specs with RT on ? If RT is so important for the next generation then how is that turning out so far with the Series S only supporting the feature in one game ? DLSS also won't help developers that will need to design their games around on what is likely to be a ~1 TFlop system with portable clock speeds so realistically there is no chance that this hypothetical machine will come close to the Series S even if it did have a better RT implementation ...
DLSS even at quality settings can already be hit or miss. DLSS at lower settings and lower resolutions is far more visually inconsistent than you imagine it to be since practically no one covers those parameters. Keeping developers on a low power mobile SoC with features like RT or DLSS is purely a pipedream or a fantasy that won't work out ...
Offering features like RT (which won't even be used) or DLSS (which adds inconsistency) doesn't help developers in any way ...
Then the obvious solution is to use simpler rendering and art pipelines if you're struggling with performance. The answer isn't to transition into more complex ray traced rendering and art pipelines on hardware that can't maintain interactive framerates even on lower resolutions ...
DLSS might work well enough in the former case if developers were looking to reach even higher output resolutions. DLSS is a lost cause in the latter case when developers will avoid using ray tracing for a much stronger system like the Series S for example so the chances of a sub-2 TFlop system in portable mode keeping up is virtually impossible. Developers are going to have to make the ultimate compromise between having simpler or more complex rendering/art pipelines if they're targeting ultraportable hardware regardless if DLSS is supported or not because reaching parity with high-end consoles simply isn't possible when they have the freedom to explore more powerful art/rendering pipelines ...
I guess it's possible. With the next Switch device, I assume Nintendo will want to go with a very power efficient design. Performance per watt and per mm will be king. I'm not sure either Turing or Ampere will fit the bill so maybe going with Nvidia's latest, especially if it's targeting gaming might be a smart idea.
It honestly wouldn't surprise if the design was still 256 shader cores like the current Switch but with a higher clock, removed RT, but plenty of tensor cores. Coupled with a modern ARM A7x CPU we'd probably be looking at a 2X improvement over the current Switch.
It honestly wouldn't surprise if the design was still 256 shader cores like the current Switch but with a higher clock, removed RT, but plenty of tensor cores. Coupled with a modern ARM A7x CPU we'd probably be looking at a 2X improvement over the current Switch.
BTW if people asked the developers in the emulation community whether it was possible for the updated system to both retain backwards compatibility and feature an incompatible architecture, they'd instantly mock at the idea of it happening. If there was a new system coming with a new and incompatible architecture then there's virtually no chance of implementing backwards compatibility ...
How exactly do they not help? You have focused on Series S and the limit of utilization of such features on that platform, but it's more than a bit premature to do that is it not? Microsoft have had to wait on AMD to roll out support for those features not just for the consoles but also PC, so devs would be more or less held back by factors outside of their control, and make due with what they have in the meantime.
But insisting RT and DLSS won't be helpful is a big reach IMO. You underestimate what developers can come up with in terms of applicable solutions to solve the very type of challenges you mention. DLSS is still a technology in relative infancy; it's made major strides from 1.0 to 2.0 and will continue to do so in the future. It wasn't that long ago that features we take for granted now like FSAA and SSR were in their infancy and were inconsistent in early implementations...but GPU designers stuck with it. Developers stuck with it. And now not only are those features very matured but the tools and techniques to leverage them have also matured across the industry, experience has been gained with them, etc.
We can't expect these things to just magically "work" out of the box; it will take time. But as features which directly bring benefits in performance costs, RT/DLSS etc. are far from being useless. It will just take more time for them to be successfully utilized by the wider slice of the industry. And those benefits will trickle down to even something like Series S, or a Switch Pro. Predicating the effectiveness of these features solely to the generic compute potential of the device doesn't seem like the best bet going forward in an industry where major developments are happening that decouple many tasks from raw compute to specialized acceleration and silicon supporting specific feature sets. Not to mention, architectures themselves having finer granularity in their configurations giving greater flexibility between raw compute, caches, and dedicated embedded acceleration.
D
Deleted member 13524
Guest
DLSS is over 2 years old.. When would you declare it to not be in its infancy? 202x?
What is cheaper to do in RT than with a rasterization "trick" counterpart?
I get adopting DLSS to improve performance, but RT? On a Nintendo handheld nonetheless?
Nintendo likes gimmicks like that. They could use RT to make some awesome lighting puzzles in their games.
And having RT acceleration is definately a good thing moving into next gen.
Again, if a feature becomes unused then how is it supposed to be helpful ? Considering most of the big developers have had a lead time of about 1 year since they've received their dev kits, it should be a cause for concern to the series S that they aren't using this feature from the early onset in this new cycle when we go by recent trends. If we take a look at the past cycle and the one before that, you'll notice that there are resolution drops from new releases overtime by developers rather than maintaining a resolution throughout the cycle. Arguably, the best time to implement and showcase ray tracing on the Series S would've been immediately in this generation where games have far simpler art/rendering pipelines but the fact that developers are finding out that they don't have the headroom left to do this suggests that the system doesn't have any further capacity to handle more general purpose or generic art/rendering pipelines with the incoming inevitable resolution drops. If these pipelines are nearly out of reach now for a less powerful system then what are the odds of an even weaker system running future games that will use more powerful and demanding pipelines ?
I never said that these features won't be helpful but it's a naive assumption that otherwise will apply in the general case as we can already see that this is not true for one current system. As for SSR, I'm pretty sure the hardware designers didn't build hardware acceleration for that technique and it's just something that was invented and implemented by graphics programmers because there were no other alternatives. Even then SSR will be largely replaced with RT reflections in the long run because it's fundamentally more consistent to begin with so expecting developers to hold back making progress on their own art/rendering pipeline by not removing these limitations for one platform is purely egotistical ...
The first paragraph above already outlined why the opposite will happen. The chances of developers being able to find enough frame time to integrate these features will start to become slimmer once resolution drops will set in later in the cycle. If they do release a hypothetical portable device with these features during this year then it will never see these benefits seeing as how the next closest system (Series S) that's struggling will be almost twice as powerful even with the highest clocks. The probability of seeing these benefits could improve if they decide to push back the release but developers still have to design their games around for portable use which will certainly see lower clocks so that's another roadblock for them to deal with ...
There exists plausible reasons why RT won't materialize on much weaker platforms for sometime such as an existing system struggling, more demanding art/rendering pipelines being introduced overtime, designing for worst case power limits and all of these combined are more than valid explanations as to why other platforms won't be able to follow high-end systems. DLSS isn't going to change this outcome either since we have an example of a low-end console already reaching it's limit early in the cycle when it's running 1080p with art/rendering pipelines that don't feature ray tracing so possibility of DLSS somehow making more room to run more generic pipelines largely unsubstantiated especially on much a less powerful system ...
I guess when a larger net of developers get to using it seriously. Which, definitely hasn't happened yet. Maybe give it 'till 2023 and it should be a much more established and understood tech.
Keep in mind too that a lot of current games are still being primarily developed for older paradigms that don't factor in DLSS from the onset, kind of the same with a lot of cross-gen games just throwing RT into the mix.
Certain AI routines, like those tied to dynamic shadows.
Measured doses of RT of course; nothing that even has to necessarily be used for what you'd typically think of RT use for. Like with the AI detection bound to dynamic RT-shadows or using RT for audio detection by the AI in certain games. Stealth games could really benefit from that type of stuff.
I'm guessing even with something like a Switch Pro/Switch 2 these uses could be doable as you don't have to use RT exclusively for them, just parts of them, and in some cases like with the audio example there's nothing from the RT actually being rasterized to the framebuffer, it's all just internal calculations using the BVH intersections to test for collisions, they don't even need to be done exclusively through RT (some of the calculations can be done with fixed prebaked values for example).
Scale of games and budgets are going to be more a determining factor than relative power.
Who said RT won't be used on Series S? It's baked into the hardware, if it's there then someone's going to use it, be it a MS developer or a 3P developer. In fact it's already in use in select games like DMC5.
Talking about Series devkits is really tricky because we know from documented fact that early devkits would not have been reflective of final silicon even going one year back because for a long time even Sony's devkits, like Oberon C0, had RT disabled because enabling it would crash the devkits (that's what I've heard of late anyway; I always though RT didn't show up because they were doing regression testing for PS4 and PS4 Pro BC...maybe both are true?).
Microsoft had to wait on AMD for finalization of RDNA 2 features and that meant Series devkits were running behind, as well as the GDK software suite stability. Devs like Capcom weren't even sure if they would be able to have RT in DMC5, literally two weeks up to its release. Before that they were seriously going to throw it in as a patch, and we're only a little over four months removed from that.
What games in particular are you referring to? The only developer of late we've heard explicitly touch on difficulty of porting games to Series S is the Control dev, and that is specifically with their title. Even that developer has said that the S is very capable, but particularly with next-gen titles as those can leverage the new RDNA 2 features. If anything, we should see more RT in future Series S titles once 8th-gen cross-gen dev has been more or less left behind.
It's not as easy as saying cross-gen games should "just work" on Series S and feature RT because the games are simpler in art/rendering pipelines because that actually isn't 100% true. A lot of the bigger AAA games from 8th gen at the tail-end have pretty demanding rendering pipelines and complex artstyles, whereas some of the titles we've seen so far targeting 9th-gen systems specifically may not be as big-budget (since they're targeting a smaller install base), and may not be as demanding. There's also the issue of certain game engines needing to be rebuilt in areas to accommodate the 9th-gen systems.
Well, I guess we'll have to see. SSR maybe wasn't the best example but there's been plenty of examples of other techniques that were invented and ran through software code on older GPUs that have since seen dedicated silicon to handle them via hardware in newer designs. ML is one such thing; it wasn't that long ago that ML models and programs targeted CPUs and simply leveraged what extended math co-processing features those had. It'd move to GPUs a bit later but even then ML-specific hardware support for things like FP16, INT8 etc. wouldn't make their way into GPU designs for a good long while.
I don't think SSR will ever fully go away; I actually am not sure if PS5 and Series X have enough capability to run full RT and provide 4K (or near 4K) rendering @ 30 FPS, let alone 60. So if we're going to see a mix of RT and SSR even on those systems as the gen wears on, we'll surely see it on Series S and Switch Pro/Switch 2. And developers will be mindful of that at the onset.
I think we need some more, concrete details on Switch Pro/Switch 2 before arriving to some of these conclusions. And we should also remember that AMD and Nvidia's architectures are in a lot of ways pretty different despite having more similarities than, say, either with Intel's Xe or Apple's stuff. We can't just look at the raw TF and go "Series S has over 2x the TF" and assume that's that. Because, we can already look at AMD's and Nvidia's current cards and see while AMD's either compete evenly or beat Nvidia (some heavily) in rasterized tasks, that's oft-times before some of Nvidia's advantages like DLSS and RT via the Tensor cores enter the picture. Currently AMD has no equivalent for those on RDNA 2; their RT is tied to the CUs and Fidelity FX is not hardware-accelerated in the way Tensor Cores are (although in Microsoft's case, their systems support DirectML which is a hardware-based solution in some manner).
For how resolution drops might impact things, I don't see how it's too much a concern as DRS can take care of it, and games leveraging whatever DLSS support Nintendo includes (probably DLSS 2.0) can operate with lowered texture quality and internal resolution which helps with frame times, use DLSS to scale the image up to a desired target resolution (it doesn't have to be 4K for the output); that saves on frame budgets and those savings can be used for any varying degree of RT. It'll just take some smart design choices but for serious devs this shouldn't be an issue after getting acclimated with things.
A lot of this, again, is predicated on judging Series S somewhat unfairly. The devkits were constantly running behind, the GDK was taking longer to stabilize, MS had to wait on AMD for certain features (some of which still aren't ready), devs needed time (some still need time) to acclimate to GDK over XDK, engines have to be retooled, porting teams may or may not have the required manpower and funds to prioritize optimizations in certain ports, etc. etc.
Features like RT aren't going to be used wholesale even on PS5 and Series X; there will be compromises there. But while you're right about rendering pipelines getting more complex, these systems have the features to accommodate that. I'm not even talking about DLSS here, but other things like Mesh Shaders, VRS (Tier 1 and Tier 2), SFS etc. . Admittedly, very Series-centric things on the console side, but Nvidia's hardware has the same features which means the Switch Pro/Switch 2 will also support them.
We have to look at these systems in the context of the whole of their capabilities, not just a single feature or two or a single metric like raw compute. Everything has to orchestrate together in order to enable absolute peak benefits. This is especially true for systems like Series S and Switch Pro/Switch 2, but once they are, things like RT will be more commonplace.
