Intel Arc Graphics A770: Windows 11 vs. Linux Benchmarks

[lyamc]

The numbers posted by Phoronix (for AMD, NVidia -- and possibly Intel -- linux vs windows) are not credible.

[snip]

Phoronix credibility is seriously under question

"I tested on my machine and the performance is pretty terrible."
"I tested on my machine and the performance is great, so you are providing false information"

That's not how it works. The performance is the performance he got. He can only report on the numbers he got. You see?

[middy]

Sorry to say but this is not a proper comparison. On the Linux side, wrappers are used for D3D11/12 to Vulkan... Not to mention any WINE overhead... It is useful as a tool to know if anyone can play those games on an Intel gpu running Linux, but is not a real comparison between the 2 drivers. You should have used games that have a native Linux version, or at the very least OpenGL/Vulkan only. I am not too familiar with Strange Brigade, but IIRC it has a Vulkan API mode, was this used in this test?

it wasn't that long ago where most of the benchmarks Michael did were native only with post after post begging him to add more proton / wine titles. now he does, its post starting to prop up asking for more native and less proton / wine titles... just can't win lol.

that said, i'm glad he's doing proton / wine games.

The issue is another one.

The numbers posted by Phoronix (for AMD, NVidia -- and possibly Intel -- linux vs windows) are not credible.
As I wrote above , my linux system goes neck on neck (if not better) with a similar windows system (AMD linux all open-source driver stack vs AMD closed source windows)
Bero Tech you tube channel (https://www.youtube.com/@beronori/videos) shows the same with his system (AMD 5800x 3d and rx6700xt)

If the problem with Phoronix is the intel E/P scheduler, Ubuntu, the lack of control in the automated scripts or something else I do not know.
What I know is that (with all respect to Michael work) Phoronix credibility is seriously under question

i don't understand why people are mad about this. especially with the amd results that was just a tiny 6.5% difference between linux and windows with the mature rdna2. you can't get any better than that when you are dealing with overhead with proton / wine. that was absolutely fantastic performance. even nvidia was very close between their linux and windows performance. again, just a small gap that doesn't make a damn difference in the end unless you care about numbers down to declaring a winner with a 0.5% difference nitpicking.

intel yes isn't that great, but its a new architecture AND intel finally get back into the dedicated gpu space since what? the late 1990s? even the windows drivers are not really that great in terms of performance and stability. its just going to take time to get some maturity. rdna3 is in a similar position right now.

sigh, i guess people won't be happy unless they see linux curb stomping windows in games meaning while i'm just going to sit here enjoying my 6900xt on linux like i have for the last two years.

[Phil995511]

Personally, I don't know if Ubuntu is really a reference OS on its own for testing games under Linux. Personally, I would prefer that you test a 100% Intel machine (CPU & GPU) under Windows 11 versus Clear Linux, Arch Linux and Ubuntu (see under Debian 11 too), in games under Steam while only using games that have been natively developed to run on Linux as well as Windows and not emulated games.

[Grinness]

"I tested on my machine and the performance is pretty terrible."
"I tested on my machine and the performance is great, so you are providing false information"

That's not how it works. The performance is the performance he got. He can only report on the numbers he got. You see?

If we talk professionalism, Phoronix should check:

* A CPU without E/P (intel, amd, whatever)
* Another Linux distribution ( and compiling a kernel with the latest patches take 10 minutes work on a 12+ threads CPU)
* run a sanity check vs well known benchmarks publishers (I looked at HW Unboxed, noticed I had the same game, same hardware - rx6800, AMD 5900x vs 58000 3D - run the game benchmark and got better results: 85 vs 81)

If we talk baby-toy stuff, then the result is a HUGE disservice to all readers and linux contributors

[TemplarGR]

And this is why I want proprietary Intel drivers for Linux!!!

AMD offers them, and they saved the month during the early Vega days.

No offense but this is the stupidest thing i read on Phoronix in quite a while...

What difference would proprietary Intel Linux drivers make? All Intel engineers are already working on the open ones... Would you have them hire another team to create Linux proprietary drivers from scratch, or perhaps just package the Windows blob and shove it into the linux kernel? What would that accomplish?

AMD's Linux proprietary drivers were always trash. I know because i used them for years... And the MESA AMD drivers were trash for many years too... The only reason radeonSI and RADV work well these days is because of the hard work Intel put into MESA even years before them, and work from others like Valve...

The issue is not the licence of the drivers, the issue is that Arc is a brand new architecture, and it is the first time in several decades that Intel attempts to make a consumer dGPU. Nvidia and AMD had decades to perfect their proprietary windows drivers. Even their own drivers were shit 2 decades ago, with many issues... And they are using a lot of hacks ("optimizations") to maximize gaming performance, "optimizations" Intel lacks because for decades their integrated gpus weren't focused on AAA gaming.

[TemplarGR]

it wasn't that long ago where most of the benchmarks Michael did were native only with post after post begging him to add more proton / wine titles. now he does, its post starting to prop up asking for more native and less proton / wine titles... just can't win lol.

that said, i'm glad he's doing proton / wine games.

I am sorry, what? First of all, i am not one of those people.

Second, i am not against testing Proton titles. They do provide significant knowledge, as i said in my previous post. The issue i have is with presenting this test run, as a way to measure the disparity between the Intel windows and Intel linux drivers. If he wanted to do that, he should had restricted his gaming pool to games not affected by Proton performance much, if at all. That means preferably linux native titles, or at the very least opengl/vulkan ones, so at least the cost of an API translator gets avoided.

[TemplarGR]

i don't understand why people are mad about this. especially with the amd results that was just a tiny 6.5% difference between linux and windows with the mature rdna2. you can't get any better than that when you are dealing with overhead with proton / wine. that was absolutely fantastic performance. even nvidia was very close between their linux and windows performance. again, just a small gap that doesn't make a damn difference in the end unless you care about numbers down to declaring a winner with a 0.5% difference nitpicking.

You don't understand because you don't want to understand it. AMD has RADV (not developed by AMD by the way...), and that Vulkan driver is more mature than Intel's ANV, but most importantly is custom tailored to run DXVK, VK3D well, since that is its primary focus. Valve is behind RADV, and they are also making Proton and financing DXVK.... So of course they have attempted to optimize their interaction much, they are even responsible for many Vulkan extensions specifically made to facilitate Proton....

It makes sense that RADV manages to close the gap considerably with Proton. Intel's ANV is a little behind.

[ms178]

If we talk professionalism, Phoronix should check:

* A CPU without E/P (intel, amd, whatever)
* Another Linux distribution ( and compiling a kernel with the latest patches take 10 minutes work on a 12+ threads CPU)
* run a sanity check vs well known benchmarks publishers (I looked at HW Unboxed, noticed I had the same game, same hardware - rx6800, AMD 5900x vs 58000 3D - run the game benchmark and got better results: 85 vs 81)

If we talk baby-toy stuff, then the result is a HUGE disservice to all readers and linux contributors

Fully agree with your list, I als agree that these kind of tests feel a bit rushed. On the other hand, Intel Arc (and even their P/E-core architecture) is still a huge beta test on Linux with ways to go. The glimpse that we've gotten today is the experience user's get when using the reviewed CPU/distro combination. Even if that distorts the overall picture a bit, that is also of value as a data point. And I am sure that additional content with Arc is in the pipeline for the future to put things into a broader context.

[Grinness]

Fully agree with your list, I als agree that these kind of tests feel a bit rushed. On the other hand, Intel Arc (and even their P/E-core architecture) is still a huge beta test on Linux with ways to go. The glimpse that we've gotten today is the experience user's get when using the reviewed CPU/distro combination. Even if that distorts the overall picture a bit, that is also of value as a data point. And I am sure that additional content with Arc is in the pipeline for the future to put things into a broader context.

I disagree on the latter.
A data point is good if it is valid. Adding noise do not add any information -- it just makes it difficult to read the 'picture' and develop an understanding.

The same CPU is used for testing Intel, AMD and NVidia linux vs Windows (and the same linux distro).
The same results across the board: linux at a fraction (sometimes 50%) of the performance of windows in gaming.
Then in the mix other workloads that somehow decrease the difference in the final geometric mean (this again makes no sense whatsoever)

This is not benchmarking, this is doing a disservice

Below the link to the previous Phoronix articles for Nvidia and AMD (radv + radonsi):

https://www.phoronix.com/review/nvid...indows11-linux
https://www.phoronix.com/review/rade...-windows-linux

[Grinness]

[...]
i don't understand why people are mad about this. especially with the amd results that was just a tiny 6.5% difference between linux and windows with the mature rdna2. you can't get any better than that when you are dealing with overhead with proton / wine. that was absolutely fantastic performance. even nvidia was very close between their linux and windows performance. again, just a small gap that doesn't make a damn difference in the end unless you care about numbers down to declaring a winner with a 0.5% difference nitpicking.

[...]

sigh, i guess people won't be happy unless they see linux curb stomping windows in games meaning while i'm just going to sit here enjoying my 6900xt on linux like i have for the last two years.

You are missing my point.
I care about quality of results. As 10+ years Phoronix reader I care about the credibility of this site
In fact I have the same issues with numbers on the NVidia Linux vs Windows article (as for AMD and Intel).

The problem is that what published by Phoronix does not align with many other sources in terms of gaming performance (see for example Bero Tech youtube channel).
All my personal tests vs published results on same GPU and equivalent CPU shows similar results Linux vs Windows (actually better, but it does not matter)

Phoronix publishes results where difference is as high as 50% Linux vs Windows
And then throws in mixed workloads (e.g. Unigine and whatnot) where Linux is similar if not stronger than Windows, reducing the difference in the final geometric mean to about 6% as you write.

This is not serious.