Question FSR 3.0 - AMD's Answer to DLSS 3.0 has landed

[psolord]

Initial tests with afmf proved underwhelming. I tested a couple of games, that had framerate issues (below 60fps) on the 6600. The driver built in performance indicator, did show the performance increase. It does not register in msi afterburner though.

The problem is not the extra performance not registering on the third party osd though. The problem is that it does not register on the screen at all. 40fps still looks and feels like 40fps although it is showing 80fps. This is in direct contrast to the fsr3 test above, which really blew my mind.

Let's see what future iterations will bring. I will try to test fsr3 on the GTX 1070 and RTX 3060ti next.

 

[Heartbreaker]

HWUB tests driver version. Unsurprisingly it's worse than the in-game version:

 

[DAPUNISHER]

I will watch Tim's breakdown later. I don't need him to know there are problems with this iteration. AFMF is a hot mess with Starfield. For context: I play games on a RX 6400, Pro 2400g and 5700G Vega iGPUs. This info should firmly establish I am not a pixel peeper. Yet AFMF is obtrusively bad in SF, to the point of ruining the experience until turned off.

 

[psolord]

I will watch Tim's breakdown later. I don't need him to know there are problems with this iteration. AFMF is a hot mess with Starfield. For context: I play games on a RX 6400, Pro 2400g and 5700G Vega iGPUs. This info should firmly establish I am not a pixel peeper. Yet AFMF is obtrusively bad in SF, to the point of ruining the experience until turned off.

Duuuude, 4GBs ain't enough! xD

 

[igor_kavinski]

Duuuude, 4GBs ain't enough! xD

AMD APUs can use up to 16GB.

https://www.reddit.com/r/Amd/comments/15t0lsm

 

[DAPUNISHER]

Duuuude, 4GBs ain't enough! xD

Hell, it isn't enough for old games like Skyrim, Fallout 4, Shadow of Mordor and War, with the texture packs. 8GB barely handles them

Let's not be pulling your personal crusade into other threads though. This is about FSR3. Which I will be trying on the 6400 at some point.

 

[Tup3x]

I will watch Tim's breakdown later. I don't need him to know there are problems with this iteration. AFMF is a hot mess with Starfield. For context: I play games on a RX 6400, Pro 2400g and 5700G Vega iGPUs. This info should firmly establish I am not a pixel peeper. Yet AFMF is obtrusively bad in SF, to the point of ruining the experience until turned off.

Well, he said the same thing. Seem to just make experience worse in current form. I don't even know why they released it like this. The only thing it improves is framerate counter (actual gaming experience be damned).

 

[DeathReborn]

Just a quick alert here for those who enable Hyper-RX or just Anti Lag/AntiLag+:

Enabling AMD Anti-Lag+ in Counter Strike 2 Will Get You Banned

With the release of AMD Software Adrenalin drivers version 23.10.1 WHQL, AMD has added Anti-Lag+ support for Counter Strike 2, but it appears that this will only result in a VAC (Valve Anti-Cheat) ban. Apparently, the Anti-Lag+ feature is implemented by detouring engine dll functions, and any...

www.techpowerup.com

I believe that AMD didn't discuss integrating AntiLag+ into CS2 and therefore it triggers VAC bans, better Dev relations could have averted this.

 

[coercitiv]

Failure is the word to describe this AMD software launch.

It has failed to convince in terms of gameplay improvements and is now further affecting AMD credibility among consumers. They launched a poorly baked package to hit an arbitrary deadline they set themselves, and the damage is arguably bigger than the benefits. Anti-lag+ may cause issues with even more games, including banning on some other multiplayer titles, though we'll need to confirm this as more reports come in.

 

[GodisanAtheist]

Failure is the word to describe this AMD software launch.

-Business as usual, then?

Like everything AMD, give it another 6 months to bake and we'll have a fairly decent thing going.

 

[dr1337]

Haven't used FSR3 but AFMF works great on my 6900xt with the beta driver. Obviously fake frames aren't great inherently but having some interpolation instead of pure frame drops in old sims and other games is really nice. Sure the input lag doesn't go away but the image remains smoother than it would have been otherwise.

And it looks way way better than any TV's interpolation I have seen, but then again its not like I'm trying to run at a constant 24fps. Staying at constant 120 when a game drops to 60-45 doesn't look the slightest bit bad to my eyes. I just wish it would run on all APIs.

 

[WelshBloke]

Failure is the word to describe this AMD software launch.

Has this actually launched? I cant find anything in the latest driver.

Still beta?

 

[GodisanAtheist]

Has this actually launched? I cant find anything in the latest driver.

Still beta?

Yep, have to download the beta driver, not integrated in the official driver branch yet.

 

[WelshBloke]

Yep, have to download the beta driver, not integrated in the official driver branch yet.

Cheers. I'll give the beta driver a go then!

 

[WelshBloke]

Ok. That straight up doubles my FPS in cyberpunk.

I'm assuming that its doing the thing that TVs do when they do that motion smoothing thing?

 

[coercitiv]

Ok. That straight up doubles my FPS in cyberpunk.

I'm assuming that its doing the thing that TVs do when they do that motion smoothing thing?

Yes, except it does in a potentially smarter way, taking frame pacing into account. When properly implemented it will be an alternative for people looking for smoother motion on fast displays. Until then it's.... beta.

It's still just motion smoothing though, and you pay for it in latency and some visual artefacts at best.

 

[WelshBloke]

I'm liking that the AMD overlay (ctrl+shft+o) has a few new entries on it in this beta driver!

 

[Ranulf]

FSR3 goes open source:


"AMD on Thursday announced the first release of FidelityFX Super Resolution 3 (FSR 3) source code through the company's GPUOpen initiative. The company just set up an FSR 3 source code repo on GitHub that game devs everywhere can take advantage of. This includes the complete source for DirectX 12, and the source of an FSR 3 Unreal Engine 5 plugin."

Via

AMD Releases FSR 3 Source Code on GPUOpen

AMD on Thursday announced the first release of FidelityFX Super Resolution 3 (FSR 3) source code through the company's GPUOpen initiative. The company just set up an FSR 3 source code repo on GitHub that game devs everywhere can take advantage of. This includes the complete source for DirectX...

www.techpowerup.com

and

AMD FidelityFX Super Resolution 3.0.3 goes open source - VideoCardz.com

AMD releases full FSR3.0 source code for DX12 and Unreal Engine 5 AMD is making the source code for FSR3.0 public today. AMD has released full source code for DirectX12 and Unreal Engine 5 implementation of its latest temporal upscaling technology called AMD FidelityFX™ Super Resolution 3 (FSR...

videocardz.com

Direct link

AMD FidelityFX™ Super Resolution 3 (FSR 3)

Discover frame generation with AMD FidelityFX™ Super Resolution 3, and get the source code and documentation!

gpuopen.com

 

[BFG10K]

FSR3 goes open source:

Another dumb move from AMD. You can't add value to your products if you make your tech available to competitors.

Imagine how much licensing fees they'd be collecting from AI vendors like NV if they hadn't stupidly given away HBM. They could've kept the best for themselves and licensed the scraps.

Now AMD doesn't get a cent, while NV has a virtual AI monopoly helped by AMD's memory tech.

GG AMD.

 

[Bigos]

What does HBM have to do with providing source to your tools?

When source code is available you can analyze what the software does without AMD involvement, saving everyone's time. It should make it more appealing to the devs.

Even now on Linux people praise the ability to tinker with AMD open source drivers. In my opinion, Steam Deck would not have been a thing without radv (which is not directly developed by AMD but is supported by AMD documentation, radeonsi/PAL source code and general mesa contributions).

Smart move by AMD.

 

[Aapje]

Another dumb move from AMD. You can't add value to your products if you make your tech available to competitors.

FSR 3 is not any better than DLSS 3, so they are not adding value, but merely not losing value by having it.

Nvidia can't really abandon their own solution and adopt FSR without losing face anyway, especially since they put so much marketing behind DLSS. So it would actually be a win for AMD if Nvidia were to adopt FSR, which they won't.

Imagine how much licensing fees they'd be collecting from AI vendors like NV if they hadn't stupidly given away HBM. They could've kept the best for themselves and licensed the scraps

Now AMD doesn't get a cent, while NV has a virtual AI monopoly helped by AMD's memory tech.

HBM was not solely developed by AMD, but also by Hynix and other partners. You have no clue whether they would have accepted a proprietary technology.

And you make the typical mistake of assuming that free technology would have succeeded if it had been proprietary, even though it being free boosts adoption. Perhaps HBM wouldn't even have been produced if it hadn't been a JEDEC standard.

GDDR6X is proprietary, but isn't actually all that different. It's just GDDR6 with a different controller that supports PAM4 signalling. The basic memory technology is the same. That is not the case for HBM. Would a company like Hynix even bother putting that much work in to just sell it to one company?

 

[SolidQ]

That was fast

also - https://steamcommunity.com/app/2375550/discussions/0/4032474633912472024/

 

[soresu]

In my opinion, Steam Deck would not have been a thing without radv

RADV and RadeonSI/AMDGPU.

Zink does exist for OpenGL, but for now at least it's not superior to RadeonSI in many places, and likely may never be given how well optimised RSI has become.

(at least unless AMD make a radical change in GPU µArch or ISA again - in which case developing a new RSI compiler backend from scratch might be too much effort ofr too little reward)

If only AMD could just run RadeonSI directly on Windows they wouldn't have OGL driver problems anymore 😭

 

[Bigos]

RADV and RadeonSI/AMDGPU.

Zink does exist for OpenGL, but for now at least it's not superior to RadeonSI in many places, and likely may never be given how well optimised RSI has become.

(at least unless AMD make a radical change in GPU µArch or ISA again - in which case developing a new RSI compiler backend from scratch might be too much effort ofr too little reward)

If only AMD could just run RadeonSI directly on Windows they wouldn't have OGL driver problems anymore 😭

Given Windows API emulation being the most important for Windows games on Linux, radv is used more than radeonsi. And indeed, the kernel drivers are shared so amdgpu deserves a praise as well. Zink is interesting, but radeonsi is so well-tuned it will be hard to surpass it as long as radeonsi development is continued (Zink can make it so that a vendor only has to develop a Vulkan driver, seems like PowerVR goes that road with their mesa driver).

Speaking of compiler backend, radeonsi currently uses the LLVM backend which is used by PAL aka amdvlk and ROCm - it is unlikely AMD would not develop such a backend for a new uarch. But radeonsi devs are eyeing using aco - currently only employed by radv - as an alternative, so it is possible in the future both radv and radeonsi will share a compiler backend, making it likely AMD will invest in it directly.

And indeed, while radeonsi is one of the best OpenGL drivers in the world, it is sad Windows AMD users are left with... whatever it is. But recently I heard AMD OpenGL Windows driver got better, maybe they are using zink or some other layered implementation instead now?

 

[soresu]

But radeonsi devs are eyeing using aco

It's already partially in the Mesa code base I think.