Ubuntu 18.10 Performance Is Looking Up, But Clear Linux Still Leads In Many Tests

[phoronix]

Phoronix: Ubuntu 18.10 Performance Is Looking Up, But Clear Linux Still Leads In Many Tests

With less than one month until Ubuntu 18.10 "Cosmic Cuttlefish" releases, I have begun my usual benchmarking dance in checking out how the Ubuntu performance is looking to its current release, in this case the Ubuntu 18.04 LTS "Bionic Beaver". Our first performance look at Ubuntu 18.10 is with a mix of seven Intel and AMD desktop systems while using Ubuntu 18.04 LTS with all updates, Ubuntu 18.10 in its current near-final form, and using Intel's Clear Linux as a gold standard reference with it generally offering the leading out-of-the-box Linux x86_64 performance of major distributions.

http://www.phoronix.com/vr.php?view=26889

[humbug]

Don't understand the headline.
Seems like there are as many regressions as there are gains for Ubuntu 18.10 vs 18.04.

[darkbasic]

How is the x265/lame thing even possible? Especially x265 which is supposed to use mostly hand-tuned assembly.

[vegabook]

Oh brother, 18.10 looks terrible for data science. What's the deal on the R and scikit-learn benches??

These two order-of-magnitude regressions smack any small percentage improvements elsewhere into oblivion.

There has to be some BLAS library shenanigans going on.

[vegabook]

How is the x265/lame thing even possible? Especially x265 which is supposed to use mostly hand-tuned assembly.

Aggressive use of AVX?

[sarfarazahmad]

You think throwing Debian testing into the mix would help since Ubuntu derives from Debian ?

[blackshard]

Aggressive use of AVX?

Yeah, and why they work in Clear Linux but not in Ubuntu?
Some compute heavy benchmarks (like h.265, but also lame encoding) are showing there is something wrong in Ubuntu 18.04/18.10, such performance gap is not sane and could not be just related to "finetuning"

[newwen]

That regresion in SciKit-Learn is very worrying. Any clue on what could be causing it?

[boxie]

That regresion in SciKit-Learn is very worrying. Any clue on what could be causing it?

Any one of a million tuning knobs set to 0.1 instead of 1.01 ?
I hope that this gets raised as a bug and fixed before release.

[Spooktra]

First of all, thanks for the x265 benchmarks, those are the most interesting encoding benchmarks to me at this point in time and second of all, what the hell is up with the x265 benchmarks? It's almost as if the x265 version tested on Ubuntu was built with a no-asm switch. A 6x faster encoding speed on one OS vs another should immediately tell you something is very wrong, I don't care what kind of "optimization" Clear Linux developers are doing, you're not going to speed up the execution of a program 6 fold just by switching to a different distro.