Ubuntu 16.10 Doesn't Change Much With Performance, Clear Linux Still Leads In Most Tests

[phoronix]

Phoronix: Ubuntu 16.10 Doesn't Change Much With Performance, Clear Linux Still Leads In Most Tests

Given yesterday's Ubuntu 16.10 final beta release ahead of the official "Yakkety Yak" debut in two weeks, I decided to run some benchmarks of Ubuntu 16.10 compared to Ubuntu 16.04.1 LTS on the same system plus also throwing in the Intel Clear Linux distribution given it tends to be one of the most performant...

http://www.phoronix.com/scan.php?page=news_item&px=Ubuntu-16.10-Beta-2-Bench

[ArthurBorsboom]

If I understand right, Clear Linux performs better mainly due to the agressive compiler flag (is that the O3?).

What is the reason the other distributions do not use this option by default?

[timofonic]

How do they achieve to use so aggressive optimIzations? Do they patch software if they find bugs? Do they merge them to upstream? Do they use GCC or ICC?

[uid313]

If I understand right, Clear Linux performs better mainly due to the agressive compiler flag (is that the O3?).

What is the reason the other distributions do not use this option by default?

I don't know. Maybe ClearLinux is compiled to only target very recent CPUs such as Skylake, hence can enable all very advanced optimizations and utilize recent instruction set extensions.
Ubuntu meanwhile have to live in the real world and support CPUs that are 10-15 year old.

[dungeon]

He, he, '-O3 -march=westmere -mtune=native blah blah' vs '-O2'

[dungeon]

And why not '-Ofast' for the win

[Aleksei]

Michael Could you throw freshly updated Arch into the pit next time?

I don't know. Maybe ClearLinux is compiled to only target very recent CPUs such as Skylake, hence can enable all very advanced optimizations and utilize recent instruction set extensions.
Ubuntu meanwhile have to live in the real world and support CPUs that are 10-15 year old.

Interestingly, their "tested supported hardware" page lists only Haswell and Broadwell and not Skylake https://clearlinux.org/documentation..._hardware.html

[doom_Oo7]

How do they achieve to use so aggressive optimIzations? Do they patch software if they find bugs? Do they merge them to upstream? Do they use GCC or ICC?

I don't know but I recompile all the software on my main system with -Ofast -march=native and I don't remember having a weird crash or something like this. -O3 may have been crashy two decades ago but nowadays it works really well.

[cj.wijtmans]

I don't know but I recompile all the software on my main system with -Ofast -march=native and I don't remember having a weird crash or something like this. -O3 may have been crashy two decades ago but nowadays it works really well.

did you build your entire system with -O3? I am generally told its not stable and gives little performance benefits.

[SaucyJack]

How do they achieve to use so aggressive optimIzations? Do they patch software if they find bugs? Do they merge them to upstream? Do they use GCC or ICC?

Dafuq? I use -03 -march=x for a lot of my software, kernel (ck kernel has this option), xorg, llvm, mesa included, and I haven't had any issues. I've even experimented with compiling more of the system that way but updates make the process too time consuming. Anyone who says -03 + -march leads to instability is spreading FUD. Probably started my Mac/Microsoft propagandists. Most(all?) benchmarks measure throughput and not latency, I hear the -march has an effect on latency but I really haven't seen solid numbers. As for raw performance, there are plenty of benchmarks on Phoronix showing what a huge difference it can make. While I haven't gone back to double check, it seemed compiling LLVM and Mesa from scratch using -03 -march had a notable effect. But I've been running that way for so long I can't say for sure.