Intel Skylake Graphics: Windows 10 vs. Ubuntu 16.04 + Latest Open-Source Driver Code

[phoronix]

Phoronix: Intel Skylake Graphics: Windows 10 vs. Ubuntu 16.04 + Latest Open-Source Driver Code

As part of the celebrations with Phoronix turning 12 years old earlier this month I ran some fun tests looking at the Windows 10 vs. Ubuntu Linux gaming performance with the new NVIDIA Pascal GPUs and also a Windows 10 vs. RadeonSI Gallium3D vs. AMDGPU-PRO comparison on the AMD side. To finish things up, here is a fresh comparison of Intel Skylake HD Graphics under Windows 10 and Ubuntu 16.04.

http://www.phoronix.com/vr.php?view=23326

[darkbasic]

Xonotic is VERY interesting here...

[schmidtbag]

This is pretty disappointing. I honestly expected much better, considering Intel's large and presumably well-paid team. Talos Principle was the only test that showed decent results (the Heaven benchmark is hardly worth noting, considering the unplayable framerates on Windows too).

[Jabberwocky]

Any idea why The Talos Principle / Metro TLL has decent mesa performance while the generic "GPUTest" tests are extremely bad compared to windows?

[bridgman]

This is pretty disappointing. I honestly expected much better, considering Intel's large and presumably well-paid team. Talos Principle was the only test that showed decent results (the Heaven benchmark is hardly worth noting, considering the unplayable framerates on Windows too).

In fairness, if you look at the corresponding AMD and NVidia Windows/Linux tests the pattern here seems to be as much "Linux game ports are slower" as "drivers are slower". There's enough information across those three reviews for some pretty interesting analysis.

[rubdos]

Xonotic is VERY interesting here...

Yes, that's indeed very peculiar, wondering why that is.

Also: this means that the AMDGPU driver stack is getting better, compared to Windows, than the Intel driver stack, compared to Windows?

http://i.imgur.com/pP9BqOR.png

Granted, the AMD stack has a bit higher deviation, so I suppose they still have to polish some stuff.

This gives an interesting view: AMD on Ubuntu is only 12% slower than on Windows, while Intel is 39% slower than on Windows. Intel's measurements are also more consistent (skew closer to 0 indicates more of a normal distribution).

I think my next computers will be AMD powered...

EDIT: Please notice that the games that ran better under AMDGPU/mesa did not get tested under Intel, so that could give a wrong picture.

[obedlink]

Off Topic Question!!
I have never played Metro Last Light Redux with RadeonSI + Mesa Git + Radeon R9 290, the games crashes at the loading screen, the PC freezes and I lost video signal, is a bug of radeonSI?

[buzzrobot]

In fairness, if you look at the corresponding AMD and NVidia Windows/Linux tests the pattern here seems to be as much "Linux game ports are slower" as "drivers are slower"...

Yes. I'd like to see tests that compare two things that are actually the same.

[LinuxID10T]

Wow, that was incredibly disappointing. I just got an i3-6100 with the HD 530 graphics and I thought they were decent. I guess they could do much better...

[slacka]

Xonotic is VERY interesting here...

Yes, I wonder why Intel struggles here. The AMD numbers are very different.

http://www.phoronix.com/scan.php?page=article&item=amdgpu-win10-fury&num=2

Is this a regression?