Windows 10 Radeon Software vs. Ubuntu 17.04 + Linux 4.12 + Mesa 17.2-dev

[anarki2]

now it is like 50% Opensource linux wins to 50% closed source windows wins.

I think in 2-3 Months the closed source windows driver is obsolete.

Yeah right, because benchmarking Linux games is soooo relevant.

[Shevchen]

If all these games were native...

Two words: Star Citizen.

[bug77]

Basically game ports suck.

Yeah, it just goes to show how bad monopolies are. Because if it wasn't for Windows' dominance, games would have been written using OpenGL. id were pretty serious about OpenGL and now they're at the forefront of Vulkan, too. But past their technical prowess, they haven't had a good game since Quake3.

[schmidtbag]

This is some fantastic progress. Though something interesting I noticed is the Linux drivers seem to perform worse as more stress is put on them. Whether that's detail level or screen resolution, it seems something is holding back potential. If all these tests were done with low or medium settings at 1080p, it seems Linux would be right on par with Windows.

Any ideas for this anomaly?

[bug77]

This is some fantastic progress. Though something interesting I noticed is the Linux drivers seem to perform worse as more stress is put on them. Whether that's detail level or screen resolution, it seems something is holding back potential. If all these tests were done with low or medium settings at 1080p, it seems Linux would be right on par with Windows.

Any ideas for this anomaly?

AMD's drivers always had more CPU overhead than those of, ahem, competition. Even on Windows. That could explain it.

[Solid State Brain]

Are these tests always OpenGL vs OpenGL or also DirectX vs OpenGL? I remember bridgman mentioning that the open source OpenGL implementation is now better than the proprietary one.

However, nobody would reasonably use OpenGL in games on Windows when DirectX rendering is available.

[bridgman]

AMD's drivers always had more CPU overhead than those of, ahem, competition. Even on Windows. That could explain it.

Actually no... the impact of CPU overhead goes down with increasing resolution, not up. You used to see this in Windows DX11 benchmarks where Fiji was (relative to competing products) faster at higher resolutions.

This looks more like either a gap in shader compiler efficiency or something we aren't doing w.r.t. efficient memory access.

[davidbepo]

very well, ubuntu 17.10 could be faster for amd gaming than windows

[bug77]

Actually no... the impact of CPU overhead goes down with increasing resolution, not up. You used to see this in Windows DX11 benchmarks where Fiji was (relative to competing products) faster at higher resolutions.

This looks more like either a gap in shader compiler efficiency or something we aren't doing w.r.t. efficient memory access.

I was thinking more along the lines of more screen real estate, more geometry to compute. But ok.
Could this also be a symptom of synchronization issues? Cause when parallelism goes up, so does the effort of keeping everything in sync.

[bridgman]

I was thinking more along the lines of more screen real estate, more geometry to compute. But ok.
Could this also be a symptom of synchronization issues? Cause when parallelism goes up, so does the effort of keeping everything in sync.

AFAIK most games keep the same geometry independent of resolution (although I guess aspect ratio might affect it a bit)... same geometry, same tesselation, just bigger triangles. That said, I believe there are a few exceptions already.

The one that surprised me was a Windows game which changed the amount of detail depending on the number of processor cores...

... which made Ryzen seem artificially slow against Intel 4-core parts