• 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.

AMD's FSR3 possibly next month ?

Man of Honour
Joined
18 Oct 2002
Posts
100,352
Location
South Coast
Not Intel, Intel already has hardware based AI upscaling for ARC cards so XeSS looks and performs better on ARC than any other GPU. This is specifically an AMD issue, always was. XeSS being used on any other card falls back to a lower level profile which isn't as precise as on ARC but is still superior to FSR. A best of both worlds approach. AMD doesn't have this luxury yet.

Isn't it self explanatory? They just need to use AI hardware based upscaling which they currently don't do as that would mean FSR being locked to AMD hardware only like DLSS and the full profile XeSS are.

 
Last edited:
Caporegime
Joined
17 Mar 2012
Posts
47,666
Location
ARC-L1, Stanton System
Not Intel, Intel already has hardware based AI upscaling for ARC cards so XeSS looks and performs better on ARC than any other GPU. This is specifically an AMD issue, always was. XeSS being used on any other card falls back to a lower level profile which isn't as precise as on ARC but is still superior to FSR. A best of both worlds approach. AMD doesn't have this luxury yet.

Isn't it self explanatory? They just need to use AI hardware based upscaling which they currently don't do as that would mean FSR being locked to AMD hardware only like DLSS and the full profile XeSS are.


RDNA 3, RX 7000 series has dedicated AI hardware, it would be right to say RDNA 2 does not, but RDNA 3 does :)
 
Last edited:
Man of Honour
Joined
18 Oct 2002
Posts
100,352
Location
South Coast
But it doesn't use it, that's the point! AMD chose not to use it for AI upscaling, whether that's because they wanted FSR to be usable by all, or because the AI HW is actually not that efficient so would be pointless (like how RTX 30 series can do DLSS frame gen as it has early gen of the hardware needed for it, but the hardware just isn't efficient like on the 40 series so would result in worse performance rather than better hence why it's not enabled for 30 series.

I suspect AMD will only enable AI hardware based upscaling for RDNA4 as all of this aligns with DirectSR being announced in Q4 and RDNA4 being readied too.
 
Caporegime
Joined
17 Mar 2012
Posts
47,666
Location
ARC-L1, Stanton System
But it doesn't use it, that's the point! AMD chose not to use it for AI upscaling, whether that's because they wanted FSR to be usable by all, or because the AI HW is actually not that efficient so would be pointless (like how RTX 30 series can do DLSS frame gen as it has early gen of the hardware needed for it, but the hardware just isn't efficient like on the 40 series so would result in worse performance rather than better hence why it's not enabled for 30 series.

I suspect AMD will only enable AI hardware based upscaling for RDNA4 as all of this aligns with DirectSR being announced in Q4 and RDNA4 being readied too.

RDNA 4 being readied in Q4, is it? Why would they not also use it for RDNA 3?
 
Last edited:
Man of Honour
Joined
18 Oct 2002
Posts
100,352
Location
South Coast
If it was efficient on RDNA 3 then they would have long since enabled it by now and gone the intel route, one profile for AMD hardware, another profile for everyone else. They did not, and this paints a certain picture as to why. They haven't even actually improved FSR2 super resolution much at all, all games that shimmered with FSR on still shimmer. They took ages to release FSR3 frame gen, and it didn't even launch with their halo game which was marketed to the eyeballs as a key AMD sponsored game. It didn't even release with the latest version of FSR2....

Also
Most leakers point to a release in the second half of 2024. One source quoted by Moore’s Law Is Dead claims that RDNA 4 will be ready to launch in the fourth quarter, but there’s a slim chance it could be released a little sooner. It’s also possible that it might be delayed until the first quarter of 2025.
Again, this aligns with DirectSR and /maybe/ RTX 50 series announcement, so AMD would be looking to at least show something around the same timeframe or do they actually prefer being behind all the time?
 
Last edited:
Caporegime
Joined
17 Mar 2012
Posts
47,666
Location
ARC-L1, Stanton System
If it was efficient on RDNA 3 then they would have long since enabled it by now and gone the intel route, one profile for AMD hardware, another profile for everyone else. They did not, and this paints a certain picture as to why.

Also

Again, this aligns with DirectSR and /maybe/ RTX 50 series announcement, so AMD would be looking to at least show something around the same timeframe or do they actually prefer being behind all the time?

Are Moores Law rumours fact now?

PyTorch, ONNX Runtime, ROCm 5.7.... was just released for RX 7000 series late last year, a year after launch of these cards.

Just because its not on the cards for launch or shortly after doesn't mean it never will, AMD's software budget is very much smaller than Nvidia's, they haven't released about 90% of software features along with the launch of the card, these things almost always come later, with the limited budget they prioritise workstation cards.
What's more they always back port any and every feature they can to older generation GPU's.
 
Soldato
Joined
30 Mar 2010
Posts
13,058
Location
Under The Stairs!
thlou_FSR3.png
 
Caporegime
Joined
4 Jun 2009
Posts
31,052

Lol.... seems another broken implementation. AMD really need to provide support from their own engineers as they should have learnt now that over the fence is never a good thing in development or at least provide a better implementation method given, which is more fool proof since this isn't just a one of now. Hopefully with the rework to use ai, they'll address this as right now, it's doing considerable damage to what reputation they had.
 
  • Haha
Reactions: TNA
Associate
Joined
19 Mar 2024
Posts
49
Location
Cybernet

Lol.... seems another broken implementation. AMD really need to provide support from their own engineers as they should have learnt now that over the fence is never a good thing in development or at least provide a better implementation method given, which is more fool proof since this isn't just a one of now. Hopefully with the rework to use ai, they'll address this as right now, it's doing considerable damage to what reputation they had.
Been messing with it on EndeavourOS Linux, not a single issue, it eats up about 7% more GPU on the 7900 XT. It makes a small difference to the base AA in a positive way.

Framegen is trash.

Radeon drivers are different on Linux so it may be a factor.
 
Last edited:
Man of Honour
Joined
18 Oct 2002
Posts
100,352
Location
South Coast
I read that AMD still play a roll. Although AMD have also said that if a game supports FSR2 then the implementation of FSR3 is much easier. It requires motion vectors from the game, so as long as there's internal TAA support, then that's what FSR uses.
 
Soldato
Joined
28 May 2007
Posts
10,071
I read that AMD still play a roll. Although AMD have also said that if a game supports FSR2 then the implementation of FSR3 is much easier. It requires motion vectors from the game, so as long as there's internal TAA support, then that's what FSR uses.
If a normal modder can do a better job then it's most likely the Dev using the tool to it's worst degree. AMD should be sending in there experts to get the best out of it if only for good publicity.
 
  • Like
Reactions: TNA
Man of Honour
Joined
18 Oct 2002
Posts
100,352
Location
South Coast
The thing is they do! Wasn't it Starfield that had AMD engineers helping Bethesda implement FSR into the engine? And even then that didn't go that well until way down the line and long after modders implemented working FSR3 and DLSS3.
 
Back
Top Bottom