DirectX12 performance is terrible on Linux

Exactly my thoughts. Many of us indeed deeply appreciate the hard work of Nvidia’s Linux team. I’m sure everyone there is accomplishing well beyond what can reasonably be expected, given the limited resources clearly allocated to Linux.

I’ve used Nvidia GPUs for 20+ years, including on systems like FreeBSD, and overall I’ve had very good—sometimes excellent—experience. There have been problems, sure, but I’ve had issues with AMD too (for example, it took several years to fix this bug).

The real problem—which I find disheartening—is the lack of transparency and communication.

It took me about two half-days to put together my own report on Starfield and Assassin’s Creed: Shadows. I didn’t expect an instant response. I fully understand there are many competing priorities—and let’s be honest, this particular issue isn’t critical. It’s frustrating, absolutely, but not on the level of GPU hangs or blank screens.

I also understand that fixing this would likely require significant developer and QA effort. If that means it takes months to address, or even longer, I get that—and I would still greatly appreciate the work.

But the complete absence of even a basic acknowledgement? That’s what’s disheartening.

I can imagine why. Big companies often have policies discouraging public acknowledgement of unresolved problems, whether due to liability concerns or PR risk. But that doesn’t make it feel any less discouraging.

And I’ll just say this, however unlikely this message is to reach Nvidia’s management:

We aren’t just gamers. Some of us work in IT, some in AI—industry or research. When we make business decisions in those places, we base them not on emotion—but on cold equations.

Trust—however—is an integral part of those equations.

16 Likes

I completely agree

2 Likes

I did exactly the same argument earlier, that in big organisations, disclosure and communication is heavily controlled. And AMD have not even set up a dedicated Linux forum!

Now saying there is absolutely no acknowledgement is not entirely true: a few tickets have been filled and @amrits has posted multiple times in this thread their filing numbers.

And I reckon that owning a 5080 I may be insulated from performance loss, because I can play at 60 fps instead of 80 with the same enjoyment.

That said your reporting was stellar. Hopefully this will be resolved this year.

4 Likes

Related Issue: DLSS: PathTracing - Reflex - Latency · Issue #292 · jp7677/dxvk-nvapi · GitHub

3 Likes

on a good light I think this dx12 issue will be solved this year, I think it might actually be resolved with driver 580, worst case scenario driver 585

I’ve seen that even if they don’t say anything it doesen’t mean that they’re not working on it, for example I’ve made a bug report here some months ago about Minecraft Big Minecraft issues with latest stable driver 565.77 on Wayland , they said absolutely nothing which was disappointing, and then all of a sudden the issue was fixed in the 575 beta driver, so there’s hope!

6 Likes

I really apologize for not communicating earlier.
Team is currently investigating performance issues, and we do have identified the root cause for Horizon Zero Dawn and are working on a fix.

32 Likes

Many thanks for the feedback.

Would you know if all these issues are game-specific, and need to be fixed on a case-per-case basis, or can we hope that when you fix it it will propagate to at least some other VKD3D games? Thanks a lot in advance.

11 Likes

Finally,some attention.

7 Likes

A fix for a performance issue found in HZD? That’s great but I found the reply very vague.

3 Likes

The difference there is that on AMD we don’t have to rely on AMD.

There we’ve got the whole stack:

AMDGPU
MESA
RADV
DXVK & VKD3D Proton + Wine
And then the game itself

I’ve seen fixes come from anybody in the above stack, where all the devs are talking to each other. We aren’t really dealing with as many AMD devs, mostly Valve Contractors & Collabora.

2 Likes

We value everything you guys do, we really do. Even if some of us get grouchy as all hell.

tumblr_2526a968c7d993aeb8d216d9c5b4fec5_1eec0a7c_540
don’t do that, don’t give me hope …

3 Likes

Thank you for this update, and for the work being done. I would like to ask, is this root cause also affecting other games?

5 Likes

@amrits I’d also like to know. Thanks for your attention on this!

3 Likes

Thank you for the update! It would be nice to hear if the root cause identified for issues with the one game is expected to improve overall performance.

3 Likes

I would also be interested in knowing if what is being investigated in Horizon Zero Dawn is more general to all games using DX12.

Despite the performance penalty, I still use a 4080 Super with Bazzite and my big OLED TV.

Here is a big benchmark video I have just published covering 20 games across windows 10, 11, Bazzite (Fedora) and Garuda (Arch) using a 4070 Super. 15% performance loss with pure rasterization, more for the VKD3D/DX12 games. Lots of data and captured footage. I have tested RT in previous videos and, yes, it’s even worse. Having said that, I can just about get a 60 fps 4K path tracing experience in Cyberpunk with the 4080S + Bazzite thanks to DLSS 4 Ultra Performance SR, though FG is very juddery so I don’t use it.

15 Likes

Nice report! Thanks.

1 Like

Thanks, this is really really good performance comparison. Some interesting results, which hopefully raise a pattern.

Interesting to see Alan Wake 2 results, as that game performed so poorly for a long time, but after some work in the drivers I noted the significant performance turn around and seems its not far off windows now.

It would be interesting to see similar comparisons on a card with less vram to see if the results are as close for some of the better performing games.

1 Like

amazing report! thank you!

1 Like

Nice with a fresh report, nicely done. Black Myth benchmark was a bit weird… I’m suprised you got such a good result in Linux, maybe because of DLSS?

1 Like