Linux vs. BSD CPU Scaling Up To 20 Threads On The Core i9 7900X

[drSeehas]

"FreeBSD 11.1 with LLVM Clang 4.0 was the fastest for compiling the Linux kernel."
YMMD

[UnholyViking]

Please redo this benchmark, if you need help with configuration of ports/packages/machine let me know and I would gladly assist.

[UnholyViking]

Because of "linuxisms" in code (be it even in the dependency) it would not compile. Some of the tests were pure bs in this regard. Why even add Linux kernel compile in the mix when you are testing also BSD's and you would know in advance that none of them would compile?

And maybe it's time to get some multi-socket (4-way or bigger) test platforms, be them even older (yeah, I know, then can be expensive). It would be more interesting than watching single socket system benches, which bring very little surprises.

I know that the vast majority of the software tested in the benchmarks actually works in BSD. Hell, I run redis and postgres in production on FreeBSD at work. It would be **very** interesting to see error logs.

[aht0]

I am curious to learn why some of these tests were not executed on the BSDs?

Because of "linuxisms" in code (be it even in the dependency) it would not compile. Some of the tests were pure bs in this regard. Why even add Linux kernel compile in the mix when you are testing also BSD's and you would know in advance that none of them would compile?

And maybe it's time to get some multi-socket (4-way or bigger) test platforms, be them even older (yeah, I know, then can be expensive). It would be more interesting than watching single socket system benches, which bring very little surprises.

[wizard69]

The OpenMPI CFD solver results show starting to hit the point that all CFD solvers eventually do of 'parallel slowdown', where the amount of inter-process communication workload has increased to the point it out-weights any gains. The gains for CFD solvers from hyperthreading are generally known to be very marginal, so 10c/20t is more likely to show that effect than 20c.

It is not uncommon for hyperthreading to be turned off on a scientific workstation. As noted sometime the gains just aren't there, sometimes you even see regressions.

[gadnet]

well same here, why does freeBSD 11 do not appear in half of the tests ?

[andreano]

I would also be interested on a comparison with Win 10

Win 10 might have a bottleneck in process destruction ATM (such as during compilation) that also affects the user interface:

https://randomascii.wordpress.com/2017/07/09/24-core-cpu-and-i-cant-move-my-mouse/

[arjan_intel]

because bsd is a joke and lost before start

So how about not doing this kind of snark?
I have great respect for anyone else who puts an OS together, be it a BSD variant or a Linux distro. Doing this is a lot of work, often thankless and all you get is bugs and complaints...
Anyone doing that deserves at least that much respect. You can think one is better than the other, and each OS will have their strengths and weaknesses, but hard work it is.

[pal666]

I can't believe you labeled this article Linux vs. BSD when most of the tests don't even have BSD in them

because bsd is a joke and lost before start

[arjan_intel]

It will be great to see glibc benchmarks with a per-thread cache enabled.

Clear Linux currently has that patch backported... and the version in these benchmarks has that.