AMD's FSR3 possibly next month ?

[MadMossy]

XESS looks identical to DLSS, If Intel can be as good as DLSS using software why can't AMD?

This is just bad, its exactly the same as FSR 2.2, what exactly did AMD do for it too look the same? I would say this isn't good enough, but its worse than that, its pathetic.

Intel is using a much higher base resolution, certainly in comparison to nvidia.

Does it need some more work, yes. But lets not forget that they have given EVERYONE frame generation, even if you aren't using FSR. Something that was previously reserved for only the 4000 series graphics cards.

 

[mrk]

XeSS in motion has image breakup, something DLSS doesn't have, FSR fizzles, XeSS stability breaks, the camera can remain static, it's moving scenery like grass and particles that appear broken up into pixel blocks - Horizon FW is a great example to showcase these or any game with foliage that waves with the wind.

 

[humbug]

Intel is using a much higher base resolution, certainly in comparison to nvidia.

Does it need some more work, yes. But lets not forget that they have given EVERYONE frame generation, even if you aren't using FSR. Something that was previously reserved for only the 4000 series graphics cards.

Is the performance gain similar? Because if not that just disqualifies the XESS result outright.

And frankly i don't care about the reasons for AMD's _____ poor upscaling tech, i only care that its _____ poor, if AMD are to have any hope of not getting to 0 - 5% market share they have to stop messing about for however long it was developing garbage tech that is indistinguishable from the previous garbage version.
AMD's ridiculousness and irrelevance in the GPU space is seemingly a hell hole with no bottom.

 

[CAT-THE-FIFTH]

Although it does make me wonder a few things:
1.)What did all the AI additions to RDNA3 bring to the table? Seems a pretty pointless uarch upgrade,if it is simply acting like a bigger RDNA2.
2.)Why did Nvidia go a route for FG,which made an RTX3090TI not run it,but an RTX4060 could?
3.)Is AMD waiting for RDNA4 to release,to actually update FSR and will it be locked to that generation? But if it needs specialised hardware,then what was the point of RDNA3?

 

[SkeeterUK]

amd says the ai stuff on rdna 3 according to their webpage about it was to assist in stuff like ingame animations and stuff, not the upscaling stuff per say but hidden stuff like that tho if anyone actually uses it ingames like that ive not heard anything.

 

[mrk]

Although it does make me wonder a few things:
1.)What did all the AI additions to RDNA3 bring to the table? Seems a pretty pointless uarch upgrade,if it is simply acting like a bigger RDNA2.

AMD originally didn't take AI/RT etc seriously, the CEO even went on the record I recall stating that it won't be that popular. They just didn't plan for it then got caught empty handed when NV ran circles around them with AI tech (DLSS) in actual games.

2.)Why did Nvidia go a route for FG,which made an RTX3090TI not run it,but an RTX4060 could?

You'd have to look back on the history of the components that frame gen uses. optical Flow Accelerators are used for this and on the RTX 20/30 they were early revisions that were not efficient which is why it was never backported to those cards. Since NV FG uses dedicated hardware (The OFAs), the next gen of OFA was needed, whch is what all 40 series cards have and they are more efficient. This was an NV dev engineering discussion on an interview a while back as well discussing it.

It would actually do more damage enabling it for older RTX cards only for people to moan that it runs like crap.

The OFA hardware is also used in many things not just for gaming frame gen, in the professional space video encoding benefits it too:

Nvidia Optical Flow Accelerators can use AI frame generation for video encoding, too

The Optical Flow Accelerators behind Nvidia's new DLSS 3 feature won't just increase video game framerates. Content creators can also use the technology to artificially increase framerates...

www.techspot.com

3.)Is AMD waiting for RDNA4 to release,to actually update FSR and will it be locked to that generation? But if it needs specialised hardware,then what was the point of RDNA3?

The last round of leaked info suggested AMD were looking at dedicated AI to do some stuff, so maybe FSR branches will require specific HW, and RDNA4 will have a gen of that HW that is capable, like how OFAs are only erformant on RTX40 and above even though RTX20/30 have them, just rubbish versions of them.

Edit*
My feeling is that AMD internally wish it never made FSR open to all. What it has done is allowed those who were on RTX30 cards, still very capable, even more capable and outshine RDNA cards further. Those 30 series owners that might have jumped ship to RDNA3 chose not to because hey, they can do frame gen after all thanks to AMD.

I don't think AMD will make the same mistake again and just like NV requires an RTX card for DLSS tech, and Intel Arc cards are needed for the full XMX quality instead of cut down versions of "universal" XeSS. If you want full quality/performance, you need proprietary HW, that is the long and short of it, and it's also why FSR still has these issues.

 

[CAT-THE-FIFTH]

I get what you are saying but RDNA3 hardware features don't seem to be utilised in any real way. Chips and Cheese did an extensive uarch overview of RDNA3 and there are decent amount of changes.

But you would design those hardware features to use them years before, when these are taped out. It doesn't seem the changes do much in the realworld and it scales like a larger RDNA2.

It just seems an utter waste of 100s of millions of USD in tape out fees alone, to just not bother. They might as well just die shrunk RDNA2 and brought forward RDNA4.

Regarding FG, I was implying it is now quite clear that you could do it via different methods. It just find it weird an RTX4060 is more powerful than an RTX3090TI. Although I suppose people sticking with older hardware is probably not what they want after the GTX1060 has lasted so long.

 

[humbug]

Although it does make me wonder a few things:
1.)What did all the AI additions to RDNA3 bring to the table? Seems a pretty pointless uarch upgrade,if it is simply acting like a bigger RDNA2.
2.)Why did Nvidia go a route for FG,which made an RTX3090TI not run it,but an RTX4060 could?
3.)Is AMD waiting for RDNA4 to release,to actually update FSR and will it be locked to that generation? But if it needs specialised hardware,then what was the point of RDNA3?

1.)What did all the AI additions to RDNA3 bring to the table? Seems a pretty pointless uarch upgrade,if it is simply acting like a bigger RDNA2.

I'm wondering that my self, its not completely useless if you're working with Zulda or ComfyUI but given those things didn't exist when RDNA 3 was being conceived the fact that RDNA 3's AI engine has a use at all is a complete fluke because its not being used in the same way it is on Nvidia cards.

2.)Why did Nvidia go a route for FG,which made an RTX3090TI not run it,but an RTX4060 could?

Because they want you to upgrade your 3090 Ti to a 4090.
I will add that AMD's strategy of offering FG to RTX 2000 and 3000 series owners as an attempt to deny Nvidia those 4000 series sales is idiotic, every reviewer: "You don't need to buy Nvidia's over priced cards if you want that but you also don't need to buy anything from AMD either"

3.)Is AMD waiting for RDNA4 to release,to actually update FSR and will it be locked to that generation? But if it needs specialised hardware,then what was the point of RDNA3?

Yeah good question, that would be really really really stupid because holding back AI generated upscaling tech is resulting in a ____ software upscaling tech being updated to an identically _____ software upscaling tech, so there's all of that bad noise and if they eventually spring AI upscaling tech on us with RDNA 4 but not for RDNA 3 enter yet another ____ storm of bad press.
This is AMD, these days i can only conclude yes, this is exactly what they have planned. Because its the most idiotic thing they could possibly do.

 

[humbug]

I'm so over this now..... its not good enough just to have great hardware, and it is great hardware, now you need supporting features, RDNA 3 has AI, with that they had one job to bring it in to line with Nvidia in those supporting features, they didn't do that job and now the AI on RDNA 3 is a waste of sand.

You had one job in this AMD.

 

[mrk]

It's not like everyone was saying exactly that for so long or anything

 

[humbug]

It's not like everyone was saying exactly that for so long or anything

They put AI on the bloody thing, its not unreasonable to think they did that for a reason.

 

[tommybhoy]

I'm so over this now.....

DLSS was available to purchase and you consciously chose not to at your pp, but you don't get that spec on Nv without cost increasing rapidly, you can't have it all.

AMD included NV/Intel FG support for those that stuck on Ampere and below for reasons, one of them was to try and disrupt 40 series main selling point-FG.

They are now running FG+DLSS=AMD have denied NV GPU sales.

NV FG could be a bust if the industry decides to concentrate on the only cross platform FG implementation that works literally the same as NV's small proprietary FG user base solution.

 

[JediFragger]

They put AI on the bloody thing, its not unreasonable to think they did that for a reason.

Iirc they use the WMMA instructions to encode/decode AV1 for streaming, my guess is that it just can't be used both for this and upscale at the same time. It's obvious now that we aren't going to get AI upscale with RDNA3 or it would have happened by now. Damn shame

 

[Joxeon]

Why did Nvidia go a route for FG,which made an RTX3090TI not run it,but an RTX4060 could?

Because the performance uplift for ADA simply wasn’t good after they cut down the dies so having FG exclusively for RTX4000 meant they could use the perf increase numbers in the marketing to make the cards look better than they were.

 

[MikeLoader]

Is the performance gain similar? Because if not that just disqualifies the XESS result outright.

And frankly i don't care about the reasons for AMD's _____ poor upscaling tech, i only care that its _____ poor, if AMD are to have any hope of not getting to 0 - 5% market share they have to stop messing about for however long it was developing garbage tech that is indistinguishable from the previous garbage version.
AMD's ridiculousness and irrelevance in the GPU space is seemingly a hell hole with no bottom.

Sure buddy

 

[humbug]

DLSS was available to purchase and you consciously chose not to at your pp, but you don't get that spec on Nv without cost increasing rapidly, you can't have it all.

AMD included NV/Intel FG support for those that stuck on Ampere and below for reasons, one of them was to try and disrupt 40 series main selling point-FG.

They are now running FG+DLSS=AMD have denied NV GPU sales.

NV FG could be a bust if the industry decides to concentrate on the only cross platform FG implementation that works literally the same as NV's small proprietary FG user base solution.

The only success AMD had with disrupting Nv tech was with Free-Sync, which is now the industry default Adaptive Sync Technology, it dominates, almost completely, even with most G-Sync branded screens they are Free-Sync tech with Nvidia branding, and that's the thing about it, Nv don't care, they have near 90% market share so the people making those screen just put Nvidia logos all over them to sell them and if you want to know if this Nvidia branded Free-Sync screen works on your AMD card... well you have to ask. Great, wonderful AMD, good job.

Pretending to be the good guy doesn't work, you're the Gazelle, not the Tiger, guess whose the Tiger?

 

[humbug]

Iirc they use the WMMA instructions to encode/decode AV1 for streaming, my guess is that it just can't be used both for this and upscale at the same time. It's obvious now that we aren't going to get AI upscale with RDNA3 or it would have happened by now. Damn shame

I use it for ConfyUI, Stable Diffusion, its pretty powerful, a 7900 XTX was equal to a 4070 Ti but those are results from early tests, ROCm has improved and the software its self has been more optimised for AMD more recently, they are faster now but after initial benchmarks people don't tend to re-run those benchmarks every few weeks, that is how fast things are moving.

AMD's AI performance is no joke, RDNA 3 are TFlop monsters, its just the software needs better support and it is getting it.

 

[humbug]

I'll put it this way.

FP8.

Nvidia H100 SXM5: 3.95 PetaFlops
AMD MI300X: 5.23 PetaFlops (absolute unit!)

 

[tommybhoy]

well you have to ask. Great, wonderful AMD, good job.

You're negating probably the greatest feat AMD's ever pulled off because there's a sticker on the box lol.

If AMD didn't create the industry standard, I wouldn't be playing games on this stunning specimen of a display as I wouldn't have paid the extra premium but more so because the other half would have bounced it out the house because of the fan.

I'm still shocked they've even tried to improve FSR-at all, never mind pull off cross vendor FG in a short space of time, I've said often enough they'd just leave it the way it was tbh.

I get what you're saying but they've only went from embarrassingly no where near Nv on a performance level to being comparatively competitive last gen-and get FSR launched is often forgotten about.

Considering dgpu is lower on AMD's priorities as it is, no way is AMD going to get within parity of DLSS quality today, Nv's got years on them with upscaling using a better solution on a higher budget.

Wondering what Nv will counter with next as the FG works.

 

[humbug]

You're negating probably the greatest feat AMD's ever pulled off because there's a sticker on the box lol.

If AMD didn't create the industry standard, I wouldn't be playing games on this stunning specimen of a display as I wouldn't have paid the extra premium but more so because the other half would have bounced it out the house because of the fan.

I'm still shocked they've even tried to improve FSR-at all, never mind pull off cross vendor FG in a short space of time, I've said often enough they'd just leave it the way it was tbh.

I get what you're saying but they've only went from embarrassingly no where near Nv on a performance level to being comparatively competitive last gen-and get FSR launched is often forgotten about.

Considering dgpu is lower on AMD's priorities as it is, no way is AMD going to get within parity of DLSS quality today, Nv's got years on them with upscaling using a better solution on a higher budget.

Wondering what Nv will counter with next as the FG works.

I will agree FG, Or AFMF is really good, that's what it is when it on green GPU's. Reviewers all acknowledge its good, i've used it and so long as you aren't violently rattling your mouse it works flawlessly, it can't do proper frame gen when its not native, but that's what FSR 3.1 FG is if you have an RDNA 3 GPU, then it works the same as DLSS FG.
---------

FSR 2.0 was released 3 years ago, 2.2 18 months ago, DLSS 3 2 years ago, so they have had 2 years to work on something as good as that, in RDNA 3, released 19 months ago they had the hardware to support it.

They have done next to nothing, at least when you compare it to AMD's own now 3 year old tech it appears that way, and they haven't taken advantage of their own hardware.

Look, CIG created their own vendor agnostic software upscaling tech, when they said last year that they were doing this what entered my mind was why? This is really weird, there is already a hardware agnostic solution in FSR.
Then it was released and i compared CIG-TSR to FSR 2.2 and thought ooooh..... that's why. FSR look's like crap.
When a game developer makes a better FSR than you. Are you even trying?

That is a genuine question, are they silently soft quitting?