Clear Linux Continues To Maintain Slight Graphics Lead Over Ubuntu 16.10

[phoronix]

Phoronix: Clear Linux Continues To Maintain Slight Graphics Lead Over Ubuntu 16.10

Back in April I did tests showing how Intel's Clear Linux distribution showed much potential for HD/Iris Graphics performance, something that intrigued many Phoronix readers since Clear Linux would generally be seen as a workstation/cloud/container-optimized Linux distribution and something with not much emphasis on the desktop or gaming. Those earlier tests were with Ubuntu 16.04, bur with Ubuntu 16.10 coming out this week, here are some fresh tests of Clear Linux and Ubuntu Yakkety Yak on an Skylake HD Graphics system.

http://www.phoronix.com/vr.php?view=23613

[dungeon]

but keep in mind Clear uses CPUFreq over P-State by default

I still keep that in mind

Maybe the best to disable totaly all those different scaling features and do one real burn-in comparison

[chrisb]

What's up with the Caffe result? Is Clear Linux using the GPU?

[Linuxhippy]

Modern compiler optimizations like profile-guided-optimizations or link-time-optimizations do make a real difference and actually pay off - firefox built with PGO loads web-pages 10-20% faster. Yes it is a lot of work, but it is really worth it - and the reason why the official firefox builds perform so much better compared to e.g. the firefox build probided by Fedora - despite using an outdated compiler (gcc 4.8.5).

I wonder how much more proof it takes until mainstream distributions like fedora and ubuntu choose to PGO/LTO build at least their low-level system-packages (XOrg, Wayland, Mesa, glib, glibc, QT, GTK, cairo, freetype, libxml, ...) instead of building everything with -O2 -fno-strict-aliasing. With SSE2 included in the amd64 instruction set by default, there is so much a compiler can do, if it is provided with additional information what the code will actually do.

[jakubo]

So maybe compiler tests would be in order. Maybe some AMD vs intel, native vs stock and all the fun things... maybe also with different schedulers...? (i cant really believe AMD sucks so hard performance-wise, yet..)

[carewolf]

Modern compiler optimizations like profile-guided-optimizations or link-time-optimizations do make a real difference and actually pay off - firefox built with PGO loads web-pages 10-20% faster. Yes it is a lot of work, but it is really worth it - and the reason why the official firefox builds perform so much better compared to e.g. the firefox build probided by Fedora - despite using an outdated compiler (gcc 4.8.5).

I wonder how much more proof it takes until mainstream distributions like fedora and ubuntu choose to PGO/LTO build at least their low-level system-packages (XOrg, Wayland, Mesa, glib, glibc, QT, GTK, cairo, freetype, libxml, ...) instead of building everything with -O2 -fno-strict-aliasing. With SSE2 included in the amd64 instruction set by default, there is so much a compiler can do, if it is provided with additional information what the code will actually do.

The biggest gain from building Qt with LTO is that the libraries become smaller, the performance difference in negligible, but the size difference is as good as building with -Os, but with -O3 performance.

[caligula]

Modern compiler optimizations like profile-guided-optimizations or link-time-optimizations do make a real difference and actually pay off - firefox built with PGO loads web-pages 10-20% faster. Yes it is a lot of work, but it is really worth it - and the reason why the official firefox builds perform so much better compared to e.g. the firefox build probided by Fedora - despite using an outdated compiler (gcc 4.8.5).

I wonder how much more proof it takes until mainstream distributions like fedora and ubuntu choose to PGO/LTO build at least their low-level system-packages (XOrg, Wayland, Mesa, glib, glibc, QT, GTK, cairo, freetype, libxml, ...) instead of building everything with -O2 -fno-strict-aliasing. With SSE2 included in the amd64 instruction set by default, there is so much a compiler can do, if it is provided with additional information what the code will actually do.

If you care about the performance, why do you care about those distros?

[Dreakon]

It may be worth looking into Solus. The project founder of Solus also works on Clear and has implemented many of the optimizations into Solus. On top of that, they've also worked hard at optimizing all of the Steam runtimes to improve game performance as well Could make for some interesting benchmarks.

[vktgz]

Clear uses CPUFreq over P-State

so intel distro does not use intel scaling driver? strange ...

Ubuntu 16.10 makes use of ... xf86-video-modesetting

and Clear uses what, I assume xf86-video-intel? so this is a comparision of different video drivers, or does intel distro also not using intel video driver?

[Jabberwocky]

The initial post over Clear Linux left me very skeptical, this time around the review/benchmark catched the attention. I still don't find the name very suitable. E.g. We have a clear advantage with the "Caffe AlexNet" test.

PS: Thanks for testing with similar mesa stacks Michael.