The Current Intel Arc Graphics Linux Gaming Performance On Linux 6.2 + Mesa 23.1-dev

[phoronix]

Phoronix: The Current Intel Arc Graphics Linux Gaming Performance On Linux 6.2 + Mesa 23.1-dev

Last week I shared my findings over the great state of Intel's open-source compute stack for Arc Graphics now that the DG2/Alchemist support was promoted to stable in Linux 6.2 and the Compute-Runtime user-space stack for OpenCL and Level Zero is back to seeing regular updates with that code having matured particularly well. Here is a brief look at the current state of the Linux gaming performance for Arc Graphics on Linux 6.2 and making use of the latest Mesa 23.1-devel OpenGL and Vulkan drivers.

https://www.phoronix.com/review/intel-arc-march23

[loganj]

for that performance it sure has a huge power consumption

[ms178]

Ouch, these gaming results hurt a lot. Especially since on Windows Arc has become more competitive these days. I wonder why Intel hasn't started sooner with this kind of Linux work. Gen 12 is out for a while now, vkd3d-proton and its importance for Linux gaming is also known. I just can think of de-priorization due to Ponte Vecchio / compute and Windows for gaming taking the priority in developer ressources.

What happened with making use of their IGC compiler within Mesa?

[mbriar]

I hope someone at Intel has time to look into that CP77 performance, because getting annihilated by a 5500XT certainly isn't a good look..

[jojo7887]

I really wish Intel to succeed with their GPU's however such poor Vulkan performance on Linux is a deal breaker.

[tildearrow]

Typo:

Valve's Portal 2 game was a simiular story where

[bridgman]

I hope someone at Intel has time to look into that CP77 performance, because getting annihilated by a 5500XT certainly isn't a good look..

Seems like the CP77 performance is CPU limited on the Intel parts, although on other games it seems GPU limited as you would expect.

[RejectModernity]

I hope someone at Intel has time to look into that CP77 performance, because getting annihilated by a 5500XT certainly isn't a good look..

Only CP77 concerns you? How about I tell you that the vast majority of the DX12 games simply don't work? This issue is year old and Intel engineers haven't fixed it. I fell for the "INTEL LOVES LINUX" meme, and I regret switching from rx580 to a770. Maybe over time they will fix it, but I fear this GPU gonna be obsolete by then. Only DX11 and Vulkan (Doom Eternal has artifacts LMAO) games are playable.

[aufkrawall]

Ouch, these gaming results hurt a lot. Especially since on Windows Arc has become more competitive these days.

But still has severe issues in important titles. E.g. Unreal Engine TSR in Fortnite ruins performance on Arc, but is fine on AMD & Nvidia. As a result, 6700 XT is 50-60% faster than the A770 when both tested with TSR high 66% (1440p Nanite & Lumen off).
It also drew 40W in idle, despite configuring ASPM as Intel recommends. Maybe next gen of their GPUs might be ok, but this one is a lost cause in the end. Real-life experience is just completely unacceptable, especially given the still relatively high prices. And 8GB of VRAM just suck anyway.

[Linuxxx]

Only CP77 concerns you? How about I tell you that the vast majority of the DX12 games simply don't work? This issue is year old and Intel engineers haven't fixed it. I fell for the "INTEL LOVES LINUX" meme, and I regret switching from rx580 to a770. Maybe over time they will fix it, but I fear this GPU gonna be obsolete by then. Only DX11 and Vulkan (Doom Eternal has artifacts LMAO) games are playable.

If you had actually bothered to read Micheal's article, then you would have noticed that not only was this already mentioned, but also pointed out towards the necessary pre-requisite to getting it solved in the first place:

For Linux gamers one of the main limitations does still, sadly, remain. As of writing there is still no sparse residency support for the ANV driver that in turn prevents many newer Windows games from working with the Intel graphics driver. The Vulkan sparse residency functionality is needed by VKD3D-Proton with Valve's Steam Play for running many modern D3D12 Feature Level 12_0 and higher games. Many modern games need this functionality that isn't yet wired up. See this ticket for more details and some of the games impacted. This just isn't contingent upon new Mesa code but first the Intel kernel driver's VM_BIND functionality needs to land, for which we'll be waiting until at least the Linux 6.4 cycle to see that happen, and then after that the Mesa code can be equipped to make use of it. So far there appears to be no public ETA for when these pieces will be wired up but hopefully it will happen this year.

That being said, I agree that Intel should have tackled this shortcoming a year ago.

Then again, the really great open-source MESA Vulkan performance out of AMDGPUs wouldn't have materialized without Valve's invalueable investments, either...