Nvidia irresistibly eclipse the list of thebest graphic cards , and that for the most part comes down to its feature set that ’s been enabled throughDLSS . AMD is n’t sitting lazily by , however . The party is explore novel means to leverage neural electronic connection to enable real - timepath tracingon AMD artwork cards — something that , up to this point , has only really been possible on Nvidia GPUs .
AMD treat the research in a blog post onGPUOpen , say that the finish is “ move towards actual - metre path tracing on RDNA GPUs . ” Nvidia already uses AI accelerators on RTX graphics cards to upscale an image via DLSS , but AMD is focused on a slightly different slant of public presentation addition — denoising .
When enabling path trace in a game likeAlan Wake 2orCyberpunk 2077,you’re only father a small fraction of the rays mold into the scene . In a veridical - time context of use , only a fistful of sample per pel are cast into the scene , and they bounce around , but rarely go back to a light germ within the scene . That run to a noisy image — see the top left of the trope above — that needs to be cleaned up with denoising . AMD is put on a neural web to the denoising process .
Nvidia has this proficiency covered already with Ray Reconstruction , which is aDLSS feature that ’s painfully underrated . It have a monumental conflict in simulacrum quality , preserving detail in path tracing that would normally take minutes or hours to render for a undivided frame offline . AMD is look at something standardized : pack a belittled numeral of sample per pixel and reconstructing the fine detail of way tracing using a neural connection .
The technique AMD is researching , however , combines upscaling and denoising into a single neural web . “ We research a Neural Supersampling and Denoising technique which generates high - quality denoised and supersampled images at eminent display firmness of purpose than render resolution for real - sentence path tracing with a undivided neural connection , ” the web log post take . “ Our proficiency can replace multiple denoisers used for dissimilar light effects in rendering railway locomotive by denoising all noise in a single pass , as well as at low resolution . ”
This take care like some foundational research for the next interlingual rendition of AMD ’s FSR , which could ultimately pit Nvidia on performance and look-alike quality . The lounge question is if these techniques expect any bespoke hardware . Nvidia claims that consecrate accelerators on its RTX graphics cards are necessary for AI - assist upscaling and denoising with DLSS , so AMD may require dedicated hardware on its GPUs , too .
However , there is a world where AMD could open up up FSR 4 — or whatever the next rendering is call — to all art visiting card while still leveraging a nervous web . RTX GPUs already have the hardware , and we ’ve see with features like Intel ’s XeSS that it ’s potential to feed AI manakin on GPUs through freestanding teaching , though usually with a hit to epitome quality and performance .
