Intel Arc Graphics A750 + A770 Linux Gaming Performance

[Anux]

coder THX

https://www.techpowerup.com/review/i...le-bar/28.html

Now I can see them atleast with internet explorer under Windows ^^ but that page shows 30% FPS loss. How does this back up rogerxs claims? And we still don't know if there is heavy stuttering like with A380 on PCIe 3 without ReBAR.
Edit: we know it's in the conclusion: unplayable stutter-fest
So no, no good upgrade for old systems.

[frustum]

I see them.

There are 3 charts, each showing relative performance at different resolutions. The baseline (100%) is A770 running @ PCIe 4.0 x16 with ReBAR. I'm putting them in table form, since there are 2 independent variables.

If you get the web page to work, you'll also get to see other graphics cards.

There is no difference between ReBAR enabled and disabled in Vulkan when there is not much traffic between CPU and GPU.

ReBAR on (PCIe 3.0) : https://gravitymark.tellusim.com/rep...1abbd1b7949fde
ReBAR off (PCIe 3.0 4x eGPU) https://gravitymark.tellusim.com/rep...45d7a25b0cacc5

But at the same time, Direct3D12 API has terrible stuttering in eGPU mode, even with small traffic.
Hopefully, the Linux driver will work not worse than the Window version.

Moreover, one of our A380 failed during the benchmark run.

[coder]

There is no difference between ReBAR enabled and disabled in Vulkan when there is not much traffic between CPU and GPU.

ReBAR on (PCIe 3.0) : https://gravitymark.tellusim.com/rep...1abbd1b7949fde
ReBAR off (PCIe 3.0 4x eGPU) https://gravitymark.tellusim.com/rep...45d7a25b0cacc5

Is that benchmark really a good characterization of Vulkan-based games?

[frustum]

Is that benchmark really a good characterization of Vulkan-based games?

The best game characterization is a game because all games use different renderers/shaders/APIs (even if they are based on the same engine).
GravityMark is a good characterization of clear GPU and driver/API performance with minimal API overhead.

[rogerx]

For those having older platforms, and upgrading their graphics card while wondering how the Intel ARCs will perform on older platforms/CPUs, I did a little research.

Comparing past NVIDIA GTX 670 (2012) and NVIDIA RTX 3060 (2021) benchmarks, assuming the Intel ARC 750/770 performance runs equivalent with the NVIDIA RTX 3060 performance. (Most benchmarks recent benchmarks show the Arc equivalent or exceeding the RTX 3060 performance.)

Guess work with Rebar performance. Recent Intel Arc reviews, those using recent motherboards/CPUs, have stated an obvious ~20% decrease in frame rates when not using PCI-E rebar function. This rate of difference may decrease or increase for older platforms, likely decrease? On the flip, the de facto graphics cards show negligible, if any, performance drops with/without Rebar, while the Intel Arc desires Rebar functions.

So instead of upgrading your NVIDIA GTX 670 on an older platform with a NVIDIA RTX 3060, which would provide a ~120% performance increase, upgrading with a Intel Arc 750/770 may only see a ~100% performance increase. For Linux users. since the drivers are open source with none of the proprietary visible artifacts, are we really loosing anything by not buying into the NVIDIA RTX 3060 or faster GPU cards? My guess, we are not. Or is this all wishful thinking.

[Guest]

I wouldn't call these a great buy. Right now the 6600 XT is faster than the A770, and is usually around $300 new. On Ebay the 6600 XT can be found used for $250 and sometimes even less. The RTX 3060 is just a little bellow $300 off Ebay used. Maybe if the A750 was $200 and the A770 was $250 I'd be all for it, but at nearly $300 you have far better choices.

I'd trade either my 6600 XT or 3060 right now for Arc. AMD drivers are boring, and I managed to finally run into one of those major driver issues with a NVIDIA driver update (black screens on Ampere with HDMI on 515.76). Intel seems like it would be interesting to mess with

[ColdDistance]

For those having older platforms, and upgrading their graphics card while wondering how the Intel ARCs will perform on older platforms/CPUs, I did a little research.

Comparing past NVIDIA GTX 670 (2012) and NVIDIA RTX 3060 (2021) benchmarks, assuming the Intel ARC 750/770 performance runs equivalent with the NVIDIA RTX 3060 performance. (Most benchmarks recent benchmarks show the Arc equivalent or exceeding the RTX 3060 performance.)

Guess work with Rebar performance. Recent Intel Arc reviews, those using recent motherboards/CPUs, have stated an obvious ~20% decrease in frame rates when not using PCI-E rebar function. This rate of difference may decrease or increase for older platforms, likely decrease? On the flip, the de facto graphics cards show negligible, if any, performance drops with/without Rebar, while the Intel Arc desires Rebar functions.

So instead of upgrading your NVIDIA GTX 670 on an older platform with a NVIDIA RTX 3060, which would provide a ~120% performance increase, upgrading with a Intel Arc 750/770 may only see a ~100% performance increase. For Linux users. since the drivers are open source with none of the proprietary visible artifacts, are we really loosing anything by not buying into the NVIDIA RTX 3060 or faster GPU cards? My guess, we are not. Or is this all wishful thinking.

I prefer to spend more money on an Intel GPU instead of buying a NVIDIA card. Linux users are used to see the operating system like a steeplechase and the NVIDIA driver is still extremely painful for desktop tasks. Yes, NVIDIA has amazing technologies, but I prefer to have a bit worse performance and get a true compatibility through the standard graphics stack.

[Dukenukemx]

I'd trade either my 6600 XT or 3060 right now for Arc. AMD drivers are boring, and I managed to finally run into one of those major driver issues with a NVIDIA driver update (black screens on Ampere with HDMI on 515.76). Intel seems like it would be interesting to mess with

AMD drivers are getting a lot of help from Valve. We may see this with Intel but right now I'm judging it based on its current performance and price. As it stands right now a 6600 XT on Linux is faster. In Windows you may find the A770 to be faster but only in DX12 and Vulkan games. Like I said, maybe if Intel dropped the price a bit more I'd go for it.

[yeeeeman]

i dont understand why they take so long with driver side

https://www.semiaccurate.com/2022/09/02/why-is-intels-gpu-program-having-problems/

TLDR: they had their driver experts in Russia and because of the war they are now in the wind...

[gsrcrxsi]

If all you care about is OpenCL compute, is Linux kernel 6.0 necessary? This seems to be an unknown yet.

would be great to have this card setup for compute on something LTS like Ubuntu 20.04 or 22.04.

I’ve installed the 6.0 kernel on one of my 22.04 systems, but it comes with its own drawbacks. (Certain things don’t work, like zenmonitor driver).

what about GPU compute performance with an older platform like E5v4 Xeons?

please someone test this.