FSR Thread

Comparison video of FSR 1, FSR 2 mod and DLSS 2.4 in Dying Light 2

 

How about some FSR 2.0 for Read Dead Redemption 2?
https://github.com/RealIndica/RDRFSR2
(Like all mods for Rockstar games, please avoid using this online as I am unsure if you will get banned.)

 

I did try the fsr2 mod for dying light 2
it's quick and easy and looks really good imo
but it introduces some weird pixel shimmer all over the image. like they're moving around a little.it's not terrible but it's there and if you're used to dlss image stability you'll see it instantly.
still really great mod overall imo

that's nice
will try
but it's already a performance hit using dx12 over vulkan

 

well I tried fsr2 mod for dying light 2

it was really promising but the ghosting is just too much. I've never seen this amount of it anywhere.

the fsr2 image is really good imo, compared to dlss image reconstruction is a little worse on fine, thin objects like tree branches and lines but the overall image perception is that fsr2 is crisper than dlss.

there is a fair bit of shimmer whereas dlss has none, however fsr2 handles camera motion better while dlss in dying light 2 is still borked in that regard.

I thought I preferred fsr2 after a few minutes, until that is you see the amount of ghosting on animations. it's everywhere - on zombies, on vegetation, on swinging ropes and particles, and it's quite bad. they need to fix it beacuse if it wasn't for it would be really,really good. Preferred to borked dlss implementation.

 

Yeah, ghosting is the problem, not sure if PotatoOfDoom can do something about it, at least he is aware of it.

 

he says he needs further work, it's all going to be based on how dlss handles information that needs to be discarded

 

Doesn't work for me, DLSS is not selectable anymore in the game's menu.

 

That's nonsense and you know it. AMD is not suddenly horrendous at DX11. It was somewhat bad for a while, in purely CPU bound scenarios (2014 - early 2016?). The devs just relied on a crutch (that being the DX11 multithreading stuff Nvidia has in their drivers) and never properly tested how the game is doing its game logic / draw calls.

That's why when played on AMD and the Nvidia driver magic isn't present, performance tanked.

 

The Lack of Exclusive Fullscreen isn't about input lag or "it has Direct flip anyway". I'm so bored and tired of having to explain this over and over to people. It's for being able to use DSR/VSR within the game. If a game DOES NOT have Exclusive Fullscreen, you need to manually adjust your desktop resolution first before starting the game and then readjust the desktop resolution back down when you exit the game.

I have a 1080p monitor and some games that don't have a native resolution slider, I straight up set in-game 1440p, 1800p, 2160p and enjoy better visuals.

 

amd have dx11 multithreading now ? I wasn't aware.
cause if not then the issue is very much the same no matter if it's dx11 from 2015 or 2022

cpu bound, 3070 is 5% faster than 6900xt
https://www.purepc.pl/test-wydajnos...orownanie-jakosci-obrazu-nvidia-dlss?page=0,8

gpu bound, 6900xt is 40% faster than 3070
https://www.purepc.pl/test-wydajnos...rownanie-jakosci-obrazu-nvidia-dlss?page=0,15

do some research before you acccuse me of nonsense

 

Except it had lots of stutter in lots of games that went away with DXVK and which Nvidia hadn't in the first place. It was a crap driver, simple as that. AMD could have improved it years ago to make situation better for their customers, but they didn't. On Linux, RadeonSI was developed by mostly 1-3 people, and it's super fast when it comes to overhead/CPU-bound fps. So AMD (billion dollar company) for years didn't want to employ even a handful of developers to make situation better for their customers. That's how things happened and nothing else. If you're not held mentally hostage by a company, you can clearly see it that way.

 

Hardware Unboxed did a CPU scaling test.

In DX12 games and using a Zen 1 or Zen+, 6 core CPU, 3070-3080-3090 were slower than a 5700 XT using the same CPU at 1080p-1440p resolutions and lower settings. A 6800-6800 XT - 6900 XT were experiencing VASTLY LESS degraded performance from the R5 1600 that was otherwise completely crippling Ampere.

In DX11, Nvidia GPUs suddenly perform better than before. Not because "hurr durr DX11 multithreading", but because Nvidia has a dedicated process in their driver that splits the workload more efficiently (forcefully?). This is not a DirectX spec. This is not something AMD has an equivalent to. This is also something that eats CPU time, so if you have an older CPU or one with fewer cores/threads, performance can get crippled again even in DX11 because all the CPU is used up. Something that would not happen with an AMD GPU.

 

Games that have stutters that get fixed with DXVK on AMD also get fixed with DXVK on Nvidia. The performance overhead might vary between games and AMD/Nvidia, but both OEMs get better frametimes with DXVK than without. So meh.

 

Can be due to missing cache/CCX hierarchy optimizations of the Nvidia driver.
Are you really surprised that AMD optimizes their own driver better for their CPUs that otherwise are not optimal for that field? Don't be then.

When you got money for such a GPU, but don't want to afford some cheap-ass Comet Lake, Rocket Lake, Alder Lake or Zen 3 6C CPU, this is self-inflicted dumbness.

This is bullshit. Applications behave as expected, but go file a complaint at MS for Nvidia allegedly violating AMD user's honor by being faster. Surely they will act and stop certifying the Nvidia driver. Hilarious...

It very likely still runs much better than AMD's old crap D3D11 before recent drivers.

Nope, lots of games run vastly better in terms of stuttering on Nvidia D3D11 vs. old AMD D3D11. Be it GTA V, Just Cause 3, Days Gone, Assassin's Creed Odyssey or whatever.

Did you have a GeForce card as of lately? I had an RX 5700 XT and 6800, and sorry, but your reality doesn't check out. It's that typical close-minded AMD Reddit cultist sermon...

 

yeah yeah yeah except in reality the exact opposite is true so your theories that underusing the cpu is better can be treated as myths and fables, compare the performance hit from going 4790k->4170 on 970 and 480
https://www.purepc.pl/radeon-rx-480-vs-geforce-gtx-970-test-na-kilku-procesorach?page=0,14
https://www.purepc.pl/radeon-rx-480-vs-geforce-gtx-970-test-na-kilku-procesorach?page=0,12

what you're implying doesn't even happen in crysis 3 on 2c/4t i5s, worse case scenario the hit is the same
https://www.purepc.pl/radeon-rx-480-vs-geforce-gtx-970-test-na-kilku-procesorach?page=0,2

tell that to 6900xt owners who got 40fps in gow until dxvk brought it back to 70 fps.

looks the same to me, lol.

astonishing level of denial.
never seen one trying to spin it so that cpu overhead is good for low tier cpus. just wow. takes imagination.

 

You keep showing me that you know very little about GPUs, how GPUs work and most importantly FSR/2. Here is yet another gaffe from you in which you mistaken a 6800xt graph from computerbase (no idea when this reviews was done so there is no context) for a 6900xt in which you are trying to post about. Astonishing levels of denial to admit to yourself that this level of conversation is way to technical for you to understand. And, your knee jerk responses to being corrected leads to why we see such a high level of incompetence in your replies. As you attempt to throw insults and name calling to cover your own missteps in this thread.

But for some context I believe Cryio was referring to the HUB video: Nvidia Has a Driver Overhead Problem, GeForce vs Radeon on Low-End CPUs

Geforce drivers do have a bottleneck when using lower end cpus that are not exhibited from Radeon GPUs as demonstrated in the review. No point explaining why "to you" because you wouldn't understand nor accept it anyway unless it favors nvidia.

This is part of the reason why FSR 2.0 is so favorable for Radeon GPUs with lower spec'd PCs. I would love to see a review using FSR 2.0 on those lower spec'd CPUs just to see what kind of delta lead Radeon would have with lower end cpus.

Be that as it may. You've demonstrated that you don't like FSR. You have no benefit in using FSR. You don't agree that FSR should exist. Insinuating your insults to anyone who uses FSR as "dumb" through out this thread only shows you as a troll. Here is just a sample of your disruptive behavior:
https://forums.guru3d.com/threads/fsr-thread.440398/page-35#post-6030707
https://forums.guru3d.com/threads/fsr-thread.440398/page-26#post-6026319
https://forums.guru3d.com/threads/fsr-thread.440398/page-26#post-6026266
https://forums.guru3d.com/threads/fsr-thread.440398/page-34#post-6030488
https://forums.guru3d.com/threads/fsr-thread.440398/page-34#post-6030455

How many more people are you going to accuse of "being in denial" when they correct your inaccurate assertions? You are just being a troll. Even after being rebuked, corrected and proven wrong all that you have done is reply in a troll like manner.

Twisting the conversation and moving the goal post. I have yet seen you show an ability to actually discuss the topic in which you are replying to. There is no insight, nor any meaningful contribution that you've added here that is worth mentioning nor debating. Therefore, you resort to posting as a troll.

It is clear that this thread is not for you. Based on your negative and disruptive posts here all you've shown is that the discussion of FSR 2.0 keeps you upset because of an AMD feature you loathe.

 

lol you really have absolutely no actual point to make except for trolling
and I already posted charts with 6900xt before you absolute buffoon

 

Your assessment of AMD's capabilities is limited, to say the least. You seem to have forgotten that AMD has been operating in the red, and losing money for many years, aka a decade, until the success of Zen processors. And it didn't happen overnight, and it took time, and several years of earnings in the black, to recoup their losses, and start to have money for R&D and the ability to hire talented programmers and engineers, and to rebuild their teams. All of this took years in planning, negotiating for TSMC time, paying out their development expenses, and giving the dev teams time to rebuild their code base from scratch.

I was with Asus, and took part of the discussion where we discussed what was wrong with our laptop lines in regards to gaming laptops. And the fruits of those discussions took 1.5+ years to manifest into products.

I'm amazed how AMD managed to get done what they have being in debt for so long. Lisa Su and her team pulled off a miracle, and are now maneuvering their tech company into a position of dominance.

The fact that the driver base is being rewritten is wonderful, and it will take time to develop and debug, and we see the fruits of this with DX11 and OpenGL performance improvements in the UWD driver releases and in recent driver packages. (OpenGL not yet in driver packages)

Sorry for jumping on you, but I felt something needed to be said.

 

and I feel like RTG (dunno if that's still a thing) will always play the second fiddle to their cpu division.
the cpu division is now raking money in, and will so with am5,now is the time their gpu division will grow at a faster rate.

this needed to be said indeed, but from a consumer point of view, we're looking for quality not excuses so he's every right to say that too.

if you were waiting to jump on amd gpus, but still felt nvidia offered a better overall product, rdna3 will be the time imo. I know I might.

 

Just dropping some meme