Phoronix benchmarks should be based on Gentoo

[enihcam]

I thought Phoronix benchmarks should be based on Gentoo, rather than Ubuntu, since Ubuntu was compiled for generic CPU and generic features whereas with Gentoo you could customize your own in make.conf.

BTW, why don't compare a customized Gentoo with Ubuntu/SuSE in various benchmarks?

[deanjo]

I thought Phoronix benchmarks should be based on Gentoo, rather than Ubuntu, since Ubuntu was compiled for generic CPU and generic features whereas with Gentoo you could customize your own in make.conf.

BTW, why don't compare a customized Gentoo with Ubuntu/SuSE in various benchmarks?

PTS is built locally on your system. When a application is to be used and has to be compiled it will be compiled to your system.

[some-guy]

PTS is built locally on your system. When a application is to be used and has to be compiled it will be compiled to your system.

But the maximum performance of it, can only match the distro it's running on, as well as the fact that it is compiled for the distro, not the computer.

@poster, I had the same opinion, but was told that these try to be real-life benchmarks, basically the average server or desktop linux, ubuntu, and fedora are mainly used.

Most computers use binary distros.

[deanjo]

But the maximum performance of it, can only match the distro it's running on, as well as the fact that it is compiled for the distro, not the computer.

@poster, I had the same opinion, but was told that these try to be real-life benchmarks, basically the average server or desktop linux, ubuntu, and fedora are mainly used.

Most computers use binary distros.

Exactly, it is a complete system benchmark including the OS. Depending what you have your build environment is set like will give you different results. Building with system defaults may give a more generic optimization but if you change your default flags a mainstream distro can test on par with a gentoo system. My default flags for example is -O3 -funroll-all-loops -ffast-math -mtune=amdfam10 -fprefetch-loop-arrays -m64 -mmmx -msse -msse2 -msse3 -msse4a -m3dnow -mabm on my openSUSE distro,

[derred]

Isn't it pointless?

Since the purpose of benchmarks is comparing results if all the tests are done in the same environment (albeit binary) shouldn't the results indicate the same thing?

[yossarianuk]

on my machine Gentoo (amd64) out performs every other OS i've tried using the phonorix-test-suite - including Ubuntu amd64 / i363 - opensuse 11.

Gentoo's etqw benchmark is about 10 fps faster...........

[energyman]

but why use ubuntu and not something like opensuse? what about a comparism between opensuse, fedora, ubuntu, SLES, RHEL?

[yossarianuk]

From my personal experience Ubuntu is slightly slower than opensuse in game/graphic benchmarks... (the only thing i'm interested in)

However from a developer point of view it is surely easier to build it for the most popular distro.........

[chithanh]

Here is a benchmark performed by the ISC which compares Gentoo against Fedora and other operating systems:

http://new.isc.org/proj/dnsperf/OStest.html

It tests only BIND 9, so the results may not be fully transferable to other areas, but it gives an idea of the performance difference.

[Vadi]

It would be one unuseful benchmark to many, the majority of the people wouldn't get the same speeds anyway.