AMD-sponsored Avatar: Frontiers of Pandora PC requirements. FSR3, DLSS and XeSS support. AMD bundles shown now

Kisai · November 6, 2023

10 hours ago, Stahlmann said:

Why shouldn't upscaling be part of the optimization process? It's just a tool same as literally any other graphics setting.

Because it's deceptive.

You know how bechmarks often turn vsync off and then try to run the game as min-maxed as possible? Well when you can't turn upscaling tech off (eg it becomes standard to have it on) then you're introducing a variable that isn't consistent across hardware.

Let's say for example, that maybe FSR produces a worse image than DLSS, but out-performs the NVidia cards in the same performance category. It becomes a race to the bottom of "how crappy can we make DLSS/FSR to outperform our competitor but users don't complain" rather than "how can we produce the best image?"

Like when you turn tunables off, you don't do it on a per-GPU level and run the same benchmark. You run the benchmark with the same settings, and FSR and DLSS are not the same, even if you can run FSR on the Nvidia GPU, you can't run the DLSS on the AMD GPU.

iSynthMan · November 6, 2023

8 hours ago, Stahlmann said:

It fits somewhere in between a 3080 and 4080, so expect 1440p high settings, DLSS quality.

But that's 3080 settings.

Could I shoot for ultra 1440 then?

RejZoR · November 7, 2023

14 hours ago, Kisai said:

Because it's deceptive.

You know how bechmarks often turn vsync off and then try to run the game as min-maxed as possible? Well when you can't turn upscaling tech off (eg it becomes standard to have it on) then you're introducing a variable that isn't consistent across hardware.

Let's say for example, that maybe FSR produces a worse image than DLSS, but out-performs the NVidia cards in the same performance category. It becomes a race to the bottom of "how crappy can we make DLSS/FSR to outperform our competitor but users don't complain" rather than "how can we produce the best image?"

Like when you turn tunables off, you don't do it on a per-GPU level and run the same benchmark. You run the benchmark with the same settings, and FSR and DLSS are not the same, even if you can run FSR on the Nvidia GPU, you can't run the DLSS on the AMD GPU.

Then we'd circle back to crappy anisotropic and texture filtering hacks that were used to gain advantage and were uncovered when reviewers started comparing image quality and not just benchmark numbers. And then it was all the rage who can make the anisotropic filtering more perfect (anyone still remembers that tube benchmark that visualized the anisotropic filtering quality at different angles and how both AMD and NVIDIA were chasing the perfect circles without any spikes shown? Yeah, if what you say happens, we're going full circle again.

Kisai · November 7, 2023

4 hours ago, RejZoR said:

Then we'd circle back to crappy anisotropic and texture filtering hacks that were used to gain advantage and were uncovered when reviewers started comparing image quality and not just benchmark numbers. And then it was all the rage who can make the anisotropic filtering more perfect (anyone still remembers that tube benchmark that visualized the anisotropic filtering quality at different angles and how both AMD and NVIDIA were chasing the perfect circles without any spikes shown? Yeah, if what you say happens, we're going full circle again.

What I'm saying is that games should not be discriminating performance tunables and then hiding them under the same one when they are in fact different.

Let's say, that LTT and GN decided to use Frontiers of Pandora as a benchmark, but FSR, DLSS, and several kinds of AA were just hidden behind "Anti-Aliasing" like a few games have done.

You'd assume that the default AA was SSAA/FSAA (super sampled, which renders the image higher and then down-samples) not FSR/DLSS (which renders the image at a lower resolution and then upsamples.) The only AA that should be used in a benchmark is "AA off" in this situation, because the underlying AA isn't exactly the same on all GPU's except for FSAA which is just downsampling. I'd probably even make the argument that benchmarks might need to turn the AA tunable off when comparing between cards of different manufacturers, since Intel's GPU's have yet a third one called XeSS. It can also change between driver versions.

However when sites like Gamers Nexus benchmarks a game, they just hit the Ultra or High preset, and that's it. There's no adjustments to the AA/DLSS settings. So if that Ultra tunable turns on FSR on AMD cards but DLSS on Nvidia cards, suddenly the performance skews heavily towards one card, or the quality skews against one manufacturer cards.

starsmine · November 8, 2023

On 11/7/2023 at 10:09 AM, Kisai said:

What I'm saying is that games should not be discriminating performance tunables and then hiding them under the same one when they are in fact different.

Let's say, that LTT and GN decided to use Frontiers of Pandora as a benchmark, but FSR, DLSS, and several kinds of AA were just hidden behind "Anti-Aliasing" like a few games have done.

You'd assume that the default AA was SSAA/FSAA (super sampled, which renders the image higher and then down-samples) not FSR/DLSS (which renders the image at a lower resolution and then upsamples.) The only AA that should be used in a benchmark is "AA off" in this situation, because the underlying AA isn't exactly the same on all GPU's except for FSAA which is just downsampling. I'd probably even make the argument that benchmarks might need to turn the AA tunable off when comparing between cards of different manufacturers, since Intel's GPU's have yet a third one called XeSS. It can also change between driver versions.

However when sites like Gamers Nexus benchmarks a game, they just hit the Ultra or High preset, and that's it. There's no adjustments to the AA/DLSS settings. So if that Ultra tunable turns on FSR on AMD cards but DLSS on Nvidia cards, suddenly the performance skews heavily towards one card, or the quality skews against one manufacturer cards.

Isn't DLSS AA, and DLSS Super Resolution two different things?

DLSS contains 4 technologies
AA
SR

FG
RR

none of which require the others to run. (but running FG on its own is dumb as fuck, so much so it should be locked out honestly). I have seen no indication that there is a significant quality or performance difference between FSR AA and DLSS AA, just SR and FG.

honestly, fuck Nvidia for putting these all in the same bucket, even if they all run in the DLSS pipeline, it's intentionally misleading the way they talk about it.
Also AMD for calling it FSR because it puts SR in the name. at least deep learning super sampling doesn't.

IDK about the argument of turning AA off, that will mislead readers about the performance these cards give you.

Also there is a fourth one with Epic's Unreal TSR

Quackers101 · November 8, 2023

8 minutes ago, starsmine said:

Isn't DLSS AA, and DLSS Super Resolution two different things?

DLSS contains 4 technologies
AA - SR - FG - RR

DLAA was it's own at first? before it was used with DLSS, not sure if that is correct. But yes it got added at some point to the mix.

starsmine · November 8, 2023

3 hours ago, Quackers101 said:

DLAA was it's own at first? before it was used with DLSS, not sure if that is correct. But yes it got added at some point to the mix.

DLSS SR I think always turns on DLAA actually so I was wrong about one not requiring the other.

But DLAA came out at the same time and You can just run DLSS AA without DLSS SR (via saying run DLSS at native resolution)

DLSS 2 added DLTAAU as well I guess.