Pemalite said:
deerox said:
Correct me if I'm wrong, but doesn't DLSS look terrible when used on a 30fps game?
|
No.
Slownenberg said:
This rumor makes it seem like DLSS will essentially be able to entirely close the gap between handheld performance and console performance. I am definitely suspicious of such a large claim. Though I do think DLSS may significantly close the gap, perhaps allowing next gen Nintendo ports of PS5/XBSeries games with performance set to lower settings, ray tracing turned off (despite this rumor claiming next gen Nintendo has ray tracing, I very much doubt that as it would be a huge waste of resources for a handheld), and minimal other changes needed. The idea, as this rumor suggests, that it will be on par with console graphics in every way, seems far fetched.
I understand DLSS 2.0 to be able to do about a 2x-3x performance enhancement. I would think that, along with running games at below 4k (after DLSS) plus 30fps for most games instead of 60fps and turning off ray tracing would probably be enough, as i said, to allow console games to play on Nintendo's handheld with no or very minimal graphical downgrades. If Switch 2 can even do this, rather than the much more significant claims of this rumor, that would be HUGE as it would mean AAA high-end graphics console games could get ported to Nintendo without much work done, just turn some settings down, turn on DLSS, and do the normal port tweaking you have to do as with any port between systems.
|
Xbox Series/Playstation 5 also have access to FSR and XeSS which are also being developed and quickly catching up to DLSS.
So the "Gap" will always exist either way.
One thing that worries me is that DLSS is propriety nVidia technology, so if Nintendo ever changes from a non-nVidia chip, then many games may loose backwards compatibility.. Which I personally believe is important going forwards as people like to bring their game libraries (Digital or otherwise) forwards.
DLSS is also used as a crutch for poorly made games, there will still be downgrades and concessions to be made as A.I. upscaling exists on the other consoles that have far more hardware resources, so the same porting issues will keep applying.
Soundwave said:
I have a suspicion that DLSS can do more than what Nvidia pushes it as because they don't want people being able to run games uses like lower power GPUs, that wouldn't be great for their sales, so they kinda really only emphasize and advertise it in areas like 8K upscaling.
They don't really push the whole "run a game at 360p and upscale it to 1080p!" aspect.
But for a device like Switch 2 ... I could totally see Nintendo basically hardwiring features like that right into the dev kit and dev pipeline. Nvidia shouldn't really care because Switch 2 is not really a threat to their GPU business, so if the dev kits for the Switch 2 are more fine tuned to utilize DLSS at super low native resolutions it's not really a big deal for Nvidia.
Also DLSS 3.5 *does* work on all RTX cards, it's just the frame reconstruction feature is only for 40 series cards, but the better image upscaling that 3.5 might provide + the ray-tracing reconstruction aspects of DLSS 3.5 will work on any card with RTX cores according to Nvidia directly. That means these features should function on the Switch 2. Who even knows maybe the frame generation feature even works. The guy who leaked the Tegra T239 chip I believe said while the Switch 2 is Ampere-based it may have some Lovelace (30 series) features too (IIRC) ... that's Nvidia 30 series cards, 30 series can do the frame generation trick.
|
Nice conspiracy, but that is far from being established on any real facts or evidence.
The fact is, the lower your base resolution, the less data that A.I. upscalers have to infer what the scene should ultimately look like at a higher resolution, the benefits of DLSS apply equally to both low-end and high-end hardware, so nVidia "holding it back" due to monetary reasons is redundant, if that was the case they wouldn't have invented the technology in the first place.
We also don't have any *real* confirmation on what chip the Switch 2 is using exactly. - So take the "leak" with a grain of salt as again it's not based on real facts or substantiated evidence.
Soundwave said:
I think what you are talking about is the frame generation feature of DLSS 3.0+ (go to 4:35 in the video):
This is frame generation, not sure the Switch 2 will feature this at all.
It can be a little confusing because DLSS is basically now 3 different things according to Nvidia
1.) The upscaling feature (taking an image of low resolution and making it appear far higher resolution)
2.) The frame generation feature was introduced in DLSS 3.0, this only works on 40 series card.
3.) Ray tracing reconstruction allows for denoising and better quality ray tracing without a performance hit.
1 and 3 should work on the Switch 2 as they work on all RTX cards. 2 though is exclusive to 30/40 series GPUs, so that may not apply to the Switch 2 at all.
|
DLSS is a group of "tools" to enhance visuals and framerates.
The same for FSR and Intel XeSS.
FSR also has "Frame Generation" and the current Switch and next-gen hypothetical Switch 2.0 will support that technology as it's platform agnostic, so even if the next-gen Nintendo console doesn't support it in DLSS, it can still have that technology from AMD... Thus it's irrelevant.
Personally I hate frame generation it gives it that soap-opera effect from a TV's motion compensation technology.
Soundwave said:
Which again is the whole point of DLSS. It will help resolve ports having to be laughably low res/muddy.
If anything DLSS feels like a technology that Nvidia accidentally developed for a hybrid console just like the Switch, lol.
Because really Nvidia doesn't want people saying "hey I can stick with my 2060 GPU and just use DLSS and not have to upgrade to a 40 series card" of course.
The tech is like a godsend for the Switch.
|
You still need to upgrade hardware, DLSS doesn't magically make low-end hardware capable of "doing everything". - I think that is an impression some people are falsely clinging to.
It's a tool.
Soundwave said:
It's gonna be a lot more interesting than I think some people might believe.
XBox Series S - 4 teraflops, no DLSS
Switch 2 - maybe like 1.25 teraflop undocked, 2.5 teraflop docked (?), but with DLSS 3.0+ (no frame generation, but super resolution scaling and ray tracing a go go)
The Series S has more horsepower naturally, but the Switch 2 can still render even as low as 360p/540p docked and get a decent enough image out of it ... the XBox Series S at 1080p has to push a lot more pixels and gets no help on the ray tracing front.
|
Series S will still have a much much much more capable CPU, faster storage, much faster memory bandwidth, FSR/XeSS (that competes with DLSS).
Using DLSS from 360P will not be a good experience, the Switch 2 will need a higher base resolution than that to look super appealing... Anyone who has used DLSS on a 480P or lower resolution and upscales it to 1440P-4k, will know that it does not even remotely look like even a 1080P image, it will have a smeared, pastey look with significant artifacts (I.E. Pixel crawl, shimmer, ghosting etc) in the rendered scene.
...Talk about over hype at the moment.
The Series S will still be ahead.
Mar1217 said:
False equivalence, the Switch which wasn't too different from the WiiU in terms of pure TFlops did manage without a sweat and that's because of the ARM NVDIA architecture. The WiiU was opting for the IBM stuff which was not up to par with x86 architectures.
But indeed, the wait and see approach is always the best
|
Switch also had:
1) Twice the memory bandwidth. (More with DCC)
2) Far better CPU performance.
3) Faster SSD.
4) More modern GPU that had Delta Colour Compression/Tiled based rendering and other efficiency gains.
It's not all about the Teraflops... Community needs to move away from that bullshit.
The Switch is more than twice as capable as the WiiU.
JimmyFantasy said:
Just to speculate a little bit further, actually we do not know anything about the Switch 2 new Dock.
This time it could include extra chips, NVIDIA tensor cores or whaterver to boost performance/visual quality for TV mode in order to have a graphics quality in line with the current gen.
I'm just speculating, but let think about it, Nintendo could even launch two Switch 2 models: one without the Dock at a lower price and a full-package including console+dock at an higher price.
|
I can tell you why the Switch 2 will not have supplemental processing in the dock.
1) Cost.
Kyuu said:
DLSS is better in practically all scenarios, but the difference on higher resolutions is minor and grossly exaggerated. DLSS might do wonders to Switch in handheld mode, but in docked mode its typical multiplats will at best be in line with Series S.
|
DLSS is far superior, correct.
But that gap is starting to narrow and FSR is gaining more industry adoption.
The best part? It runs on almost everything.
sc94597 said:
One of the biggest use-cases for super-sampling on PC handhelds (using FSR) is to save battery life, and to improve image-quality while using less GPU power. DLSS is even better at this than FSR.
DLSS 2.0 is not some marginal feature, but rather the core of Nvidia's "DLSS" feature-suite. The fact that DLSS 3.0 likely won't be part of Switch's feature-set (because of the Ampere architecture more than anything else) doesn't mean that the features included are "useless."
By the way, I am a Data Scientist who has implemented CNN's (Convolutional Neural Networks) in both the workplace, academic setting (I have a MSDS) and personal projects. I know how DLSS works on a much deeper level than most people, given that I've actually implemented convolutional autoencoders using both PyTorch and Tensorflow.
In fact I am planning to start a doctorate (D.Eng) next year, and the thesis I am leaning towards likely will incorporate CNN's in comparison to Bayesian vector time-series analyses of a certain sub-set of business problems.
|
Much of what you touched on is likely out of reach of most peoples understanding.
|
