-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
DLSS Momentum Continues: 50 Released and Upcoming DLSS 3 Games, Over 250 DLSS Games and Creative Apps Available Now
- Thread starter Nexus18
- Start date
I think you meant nvidia, Why would AMD care if you can use DLSS with FG on a 3080?amd won't be happy
Read the comment by puredark.....
AMD are blocking other "upscaling" tech being used with their frame gen, so much for "giving people the choice" eh but nope, amd decide what is best for you.
Not that it'll make any difference to me or others who want a smooth and visually good experience anyway unless puredark can sort out the issues with amds frame gen.
its all open source that's what modders do. DLSS + FG on a 3080 (if it worked properly) would only impact nvidia as FG is exclusive to the 4000 series. I hope he does get it working and that it works better than the built in FSR + FG.
Intel details its own frame generation
wccftech.com
Intel Frame Generation Technology For XeSS Could Be Coming Soon: ExtraSS With Frame Extrapolation To Boost Game FPS
Intel could be the third major PC player to debut its frame generation technology for its XeSS framework known as ExtraSS.
wccftech.com
Interesting spin on what Nvidia and AMD are doing - since frame extrapolation only uses previous frames and doesn't have to delay image rendering the way FSR and DLSS does, it could actually have less latency penalty at the cost of less precision/more artifacts.Intel details its own frame generation
Intel Frame Generation Technology For XeSS Could Be Coming Soon: ExtraSS With Frame Extrapolation To Boost Game FPS
Intel could be the third major PC player to debut its frame generation technology for its XeSS framework known as ExtraSS.
I like that each vendor has different pros and cons to their methods - gives more choice to gamers and should create more competition.
Last edited:
Hopefully intels option will actually be usable where it is a viable option, they have proven themselves with XESS and their proper approach to these techniques are very good though so I have faith in intel!
Man of Honour
Intel are now working on Frame Extrapolation, better than Frame Generation, so as Alex says, probably expect NV/AMD to adopt this technique and see current DLSS/FSR be updated using techniques Intel bring to the table for even better gains etc.
Someone's already got FSR3 FG + DLSS working
.I've tested it and works well on the 3080FE.
Also got it to work on The Witcher 3 and Ark Survival Ascended. Amazing stuff.
Last edited:
Man of Honour
It's in the very quote you posted:
Examples need to be shown of course, but it's pretty black and white in its function. Faster, better quality. XeSS is already really high quality and in various games like RoboCop, is actually temporally more stable than even DLSS.
Last edited:
You snipped the important part:
They're saying their method has better qualities than prior extrapolation methods (which are inferior to interpolation) - hence this'll have better latency than Nvidia/AMD at the cost of image quality (since guessing what the next frame is going to be isn't going to be as accurate as interpolating between two already generated frames).
Man of Honour
I think we're discussing the same thing but reading it differently.
There is no extrapolation currently, it's all interpolation, Inte's method uses future frame data with no hold as Alex describes from their take on the news.
What intel say in the quote you posted again states that because of the difficulty in the disoccluded areas of the frame, they devised a way to mitigate that difficulty by using frame extrapolation (this new "warping method"). As a result this extrapolation has better qualities than existing frame generation methods.
It's the warping method to extrapolate future frames this way which is what's giving the boost seemingly. It's aligned with Ray reconstruction where Nvidia replace several RT traditional denoisers with a single AI pass to predict what the light and ray bounces are doing which then results in a faster and at the same time, high quality image render.
There is no extrapolation currently, it's all interpolation, Inte's method uses future frame data with no hold as Alex describes from their take on the news.
What intel say in the quote you posted again states that because of the difficulty in the disoccluded areas of the frame, they devised a way to mitigate that difficulty by using frame extrapolation (this new "warping method"). As a result this extrapolation has better qualities than existing frame generation methods.
It's the warping method to extrapolate future frames this way which is what's giving the boost seemingly. It's aligned with Ray reconstruction where Nvidia replace several RT traditional denoisers with a single AI pass to predict what the light and ray bounces are doing which then results in a faster and at the same time, high quality image render.
Last edited:
There's no extrapolation on consumer GPUs currently - Intel is discussing methods of temporal supersampling of which extrapolation is one of them:
mrk said:
You're misinterpreting the article - as we can see above, they're saying their extrapolation method has better qualities than these prior extrapolation methods.
Man of Honour
The quote literally says though:
Let's just wait and see when the presentation hits the streams.
Nowhere does it say previous extrapolation techniques, this is based on current techniques, which is interpolation, which is the exact takeaway Digital Foundry made from the announcement too.
Let's just wait and see when the presentation hits the streams.
Extrapolation = Frame Generation - the whole paper is about improvements over prior extrapolation methods and interpolation (Nvidia/AMD) is only mentioned as an alternate method of Frame Generation (of generally superior quality to existing extrapolation methods whilst having greater latency).
Direct copy/paste from the paper itself (ExtraSS: A Framework for Joint Spatial Super Sampling andFrame Extrapolation):
They're proposing a method of doing upscaling (spatial SS) and frame extrapolation (temporal SS) and it's better than prior extrapolation methods.
Last edited:
Man of Honour
No there is nothing that points to the image quality being lower than FG interpolated frames currently. Direct from Intel:
Edit* Yeah to clarify, what the Intel paper says is as a baseline, frame extrapolation is lower quality than interpolation, which is why NV and AMD use latency reduction tech to supplement interpolated frame gen - Intel has worked around that by using the new warping method (mentioned earlier) to mitigate the quality and latency issue using this one method. So the end result of ExtraSS is high quality FG but without the latency inducing output requiring Reflex etc to reduce input latency. So the lower quality trait of extrapolation doesn't apply, it's just a reference knowledge point to describe what extrapolation was like, before the new warping method was introduced.
If that works across the board like XeSS generally does regardless of GPU vendor (assuming it just uses AI cores on whatever GPU is installed instead of profile specific to ARC GPUs) then for a short time it's chess move check for NV and AMD until they update FSR3 and DLSS FG to also use these new methods.
I remain aligned with Digital Foundry's take on this.
Edit* Yeah to clarify, what the Intel paper says is as a baseline, frame extrapolation is lower quality than interpolation, which is why NV and AMD use latency reduction tech to supplement interpolated frame gen - Intel has worked around that by using the new warping method (mentioned earlier) to mitigate the quality and latency issue using this one method. So the end result of ExtraSS is high quality FG but without the latency inducing output requiring Reflex etc to reduce input latency. So the lower quality trait of extrapolation doesn't apply, it's just a reference knowledge point to describe what extrapolation was like, before the new warping method was introduced.
If that works across the board like XeSS generally does regardless of GPU vendor (assuming it just uses AI cores on whatever GPU is installed instead of profile specific to ARC GPUs) then for a short time it's chess move check for NV and AMD until they update FSR3 and DLSS FG to also use these new methods.
Last edited:
I'm afraid you're going to be disappointed then. This research is interesting, but it isn't magic.
Nvidia has two perfect frames to interpolate between in addition to motion vectors and whatever buffers the engine exposes for DLSS and frame-gen - and even then, it's far from perfect. AMD has to turn frame-gen off altogether for UI elements as it's unable to resolve them.
Intel's proposed method *is* novel - and if they're able to get anywhere close to the quality of Nvidia/AMD's methods then I'll be impressed, but they're working with significantly less image data and there will be trade-offs.
Man of Honour
People said the same about Ray Reconstruction with DLSS 3.5 too, but here we are, higher quality ray tracing and reflections with zero trade-offs in performance or image quality. The by-product being about 4-5fps gained too which wasn't even part of the design process of RR but just an observed gain thanks to the efficient nature of RR.
Ok ray reconstruction relies on tensor cores to work since it uses the DLSS pipeline, but it looks like ExtraSS works similarly, that it relies on XeSS as its foundation and then introduces the new addons.
Ok ray reconstruction relies on tensor cores to work since it uses the DLSS pipeline, but it looks like ExtraSS works similarly, that it relies on XeSS as its foundation and then introduces the new addons.