Old puter - still good enuff till I save some pennies!
Because the PC Market is also quite large and does not pale in comparison to consoles and Nvidia GPUs are still the overwhelming majority (maybe not RTX but certainly mindshare). And it is also very likely that implementing Gameworks and therefore DLSS does provide some kickback from Nvidia for use and branding.
Old puter - still good enuff till I save some pennies!
I downloaded the rift breaker demo and FSR works on my Nvidia gtx 970!
There's a YouTube video in Russian where he tests a couple of games with the 970 using FSR, (nearly) free performance win win for me.
CAT-THE-FIFTH (24-06-2021),Xlucine (24-06-2021)
It seems DF TAAU testing was also flawed:
https://www.reddit.com/r/Amd/comment..._mistake_with/
From that thread:Digital Foundry made a critical mistake with their Kingshunt FSR Testing - TAAU apparently disables Depth of Field. Depth of Field causes the character model to look blurry even at Native settings (no upscaling)
Native
TAAU
Once you disable DoF for FSR Ultra Quality:
TAAU
On OcUK forums someone made a sceenshot of the DF review.
It shows FSR looks closer to the native image,and the TAAU image looks much sharper.
Apparently also AMD isn't putting a lot of sharpening into FSR images,ie,using tools such as Radeon CAS helps out.
Last edited by CAT-THE-FIFTH; 24-06-2021 at 03:38 PM.
It's a better upscaler then most of the other ones, but as the 1080p images show it's still pretty limited. The 4k images work better because the source textures aren't good enough in the games shown so are effectively already upscaled in native 4k - hence using FSR is just another more efficient way of doing the same thing.
It seems there are still issues with the DF coverage:
https://twitter.com/JirayD/status/1408383962498613251#
Even though DLSS is better,both Gamersnexus and Hardware Unboxed showed the same issue. DLSS at 4K is noticeably better than at lower resolutions,as there is less information to work with. Another issue is analysing just still images - because issues such as trailing/ghosting affected DLSS1.0 and TAAU more than say DLSS2.0/2.2 and apparently FSR.
Last edited by CAT-THE-FIFTH; 25-06-2021 at 03:12 PM.
There is the same % of information to work with upscaling from 1080p or 4k as they both start from a lower resolution containing the same % of pixels vs the native one. The difference is the 1080p image has more information in it per pixel because the source data provides more information. For example if you use a 2k texture at 1080p then you can effectively supersample AA it for a as sharp as possible an image as 1080p can give you. If you use that same texture at 4k it's already past it's limits and being internally upscaled so the texture quality per pixel won't be as good - you'd need a 8k texture for that. Hence upscaling will show less loss of information because effectively there wasn't as much to start with, hence it doesn't feel as blurry.
Like you say this is true for all upscalers including DLSS, but DLSS is better because it's got more processing power behind it (all those tensor cores) so is smarter and it's temporal so has several frames worth of data to work with not just one.
Temporal bit helps certainly, as shown in Epic's own TSR, but the bolded bit isn't the case. DLSS is better because it's reconstructing content - this isn't really much to do with processing power, but instead the model based 'if I start with these bits of information then my training suggests this is the correct reconstruction'.
Hardware Unboxed tested FSR on IGPs and older GPUs with DOTA2:
https://www.youtube.com/watch?v=WlxHsuZP2y0
Basically because DOTA is not very taxing,and like most upscalers 1080p isn't the ideal resolution in terms of image quality,it seems the game really does not need FSR at 1080p. However,it seems to make the most sense at 4K and maybe 1440p if you want decent FPS.
There are currently 1 users browsing this thread. (0 members and 1 guests)