Wine 3.10 vs. Ubuntu 18.04 vs. Windows 10 Desktop Performance

[Michael]

I think that benchmark design is again bad, why people are using Wine? Its for Photoshop, MS Office and Games, so would be nice to see such benchmarks.. we dont new Linux ports of such software we need Windows vs. Wine on Linux and Wine on Mac for some test case would be nice..

Just for fan try to ad some Android x86 wine benchmarks - to point out, that almost nothing is working in this Wine port. BTW any Android or Android x86 benchmarks would be nice anyway.

I guess you didn't read the text of the article to understand the point of these benchmarks.

[davibu]

Thank you so much!
This is really interesting.

[Grinch]

Did you compile the window binaries as well ? Otherwise the windows native vs ubuntu native are rather pointless as a comparison between operating systems, as it will be the different compiler and likely different compiler optimization setting which result in the larger part of any differences.

In particular the x264 test result stands out as I've used it extensively on both Linux and Windows and the latter has always been slower, but as I recall the official windows binary is built using PGO to help speed up those few parts that haven't been written in hand-tuned assembly, (you can of course use PGO on Linux as well, but this is not done here judging by the optimization options listed. That said, a PGO compile should not result in as much as 9% improvement on x264, so this might be another case of Ubuntu 18.04 showing subpar performance.

[sdack]

Nice. I hope Michael includes DXVK and CSMT-enabled/disabled in the Wine 3D benchmarks.

While I get the notion is there little point to it. DXVK only accelerates DirectX11 and the outcome would still be for Windows 10 to win. DXVK is fast, but it's not going to tell you anything you don't already know. Worst case would be you see a few benchmarks empty, because of hangs. CSMT may be interesting for 3D applications, but it's also come a long way and has been set to on by default now. Unless it's known to still cause major issues would a test of the option only confirm what the WINE devs already know and why it's default.

What I haven't see in the benchmarks is the mentioning of WINEDEBUG=-all which can be a significant burden for some applications as these end up spilling out megabytes of log information during a run. Would be nice to know if this was left untouched or if it was included in the benchmarks.

[duby229]

When you see numbers like that, you know it must be context switching. I mean really, it's plainly obvious.

[duby229]

Of course there can be regressions, but those benchmarks clearly show that wine is not an emulator and that a performance-wise superior underlying system can pass-through its performance even in undesigned-for usecases.

Wine definitely is an emulator. Almost nothing it does is direct, and all the while it doesn't care at all about runtime behavior. So although it is by every definition an emulator the one thing it needs to care about is the one thing it doesn't give a shit about.

[Weasel]

Wine definitely is an emulator. Almost nothing it does is direct, and all the while it doesn't care at all about runtime behavior. So although it is by every definition an emulator the one thing it needs to care about is the one thing it doesn't give a shit about.

Wine is not an emulator.

Did you know OpenGL and Vulkan aren't "direct" either, they are layers on top of the drivers which is on top of the hardware? Vulkan is an emulator, confirmed.

Adding an extra layer doesn't make it "an emulator".

[duby229]

While I get the notion is there little point to it. DXVK only accelerates DirectX11 and the outcome would still be for Windows 10 to win. DXVK is fast, but it's not going to tell you anything you don't already know. Worst case would be you see a few benchmarks empty, because of hangs. CSMT may be interesting for 3D applications, but it's also come a long way and has been set to on by default now. Unless it's known to still cause major issues would a test of the option only confirm what the WINE devs already know and why it's default.

What I haven't see in the benchmarks is the mentioning of WINEDEBUG=-all which can be a significant burden for some applications as these end up spilling out megabytes of log information during a run. Would be nice to know if this was left untouched or if it was included in the benchmarks.

Yes CSMT causes major issus. It causes massive amounts context switching. It forces every core to constantly flush and refill over and over. When you turn on CSMT and you see every core hit 100% CPU use, that's not because those cores are busy, instead it's because they are -NOT- busy at all and only flushing and refilling. Just because CSMT thrashes your CPU doesn't mean its doing anything useful, its actually not.

[Apopas]

Why is x264 encoding faster on windows?


...

[Weasel]

Yes CSMT causes major issus. It causes massive amounts context switching. It forces every core to constantly flush and refill over and over. When you turn on CSMT and you see every core hit 100% CPU use, that's not because those cores are busy, instead it's because they are -NOT- busy at all and only flushing and refilling. Just because CSMT thrashes your CPU doesn't mean its doing anything useful, its actually not.

No it's because it uses busy waiting for performance reasons.