Benchmarking OpenMandriva Lx 4.2 With Its AMD Zen Optimized Build

[phoronix]

Phoronix: Benchmarking OpenMandriva Lx 4.2 With Its AMD Zen Optimized Build

While Intel has Clear Linux as an aggressively optimized Linux distribution catering towards their hardware, there isn't a direct equivalent for optimally showcasing the performance potential of current AMD platforms. Clear Linux often offers leading performance on Zen CPUs but that is obviously not by design but just an artifact of a lot in common between the latest Intel and AMD microarchitecture features. One of the few distributions (or only notable one) offering specific AMD Zen optimized builds has been OpenMandriva. With the OpenMandriva Lx 4.2 release candidate shipping this week, I ran some fresh benchmarks looking at how OpenMandriva Lx 4.2 RC1 compares between its generic x86_64 image and that of the Zen optimized build as well as in turn how that performance compares to Clear Linux and Ubuntu 20.10.

http://www.phoronix.com/vr.php?view=29826

[dimko]

For someone on Gentoo - interesting figures.

[xpris]

Good tests. Worth to add that Clear Linux by default use Performance CPU governor while OpenMandriva only ondenamed.

[CochainComplex]

The default Clear Linux $CFLAGS flags (withouth considering avx, etc enabled special builds) are

-g -O3 -feliminate-unused-debug-types -pipe -Wall -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector --param=ssp-buffer-size=32 -Wformat -Wformat-security -m64 -fasynchronous-unwind-tables -Wp,-D_REENTRANT -ftree-loop-distribute-patterns -Wl,-z -Wl,now -Wl,-z -Wl,relro -fno-semantic-interposition -ffat-lto-objects -fno-trapping-math -Wl,-sort-common -Wl,--enable-new-dtags -mtune=skylake -Wa,-mbranches-within-32B-boundaries

according to
https://community.clearlinux.org/t/w...he-kernel/3422

edit: But I don't know if mtune=skylake is here a dynamic flag set during distro installation. In a lot of cases march=haswell is std for a lot of pkgs

[nuetzel]

LTO + PGO is needed. (Intel do some of it.)

radeonsi Mesa reenabled 'sisched' patch coming, soon. (at least for Polaris)

BTW
KDE Konqui do not work any longer with the forum software...
...so sad. --- Have to use Firefox for the first time.
(Problem started in November/December.)

[berolinux]

Clear Linux has a few modifications that help it get ahead in benchmarks and in a few real world situations, but that really shouldn't be done in a more general purpose distribution.
One example is setting the CPU scheduling governor to performance - this will always result in better benchmark results, but you certainly don't want that enabled on a laptop while on battery.

[nvaert1986]

For someone on Gentoo - interesting figures.

As somebody running Gentoo on a daily basis, I can tell you these numbers can actually be higher, depending on how you tweak and tune your kernel. If you apply certain Clear Linux patches, set the CPU Govenor on Performance (desirable on workstation, but undesirable on laptops) and use a different scheduler from BMQ (perhaps CFQ, but MuQSS or some other is recommended depending on what your main usage is), this can easily be 10 - 15%, sometimes even up to 17 or 18% and if you have a lot of machines, from which you use 1 to compile your packages on and provide them as binaries to a lot of other machines with similar hardware a 10 - 15% increase easily saves in hardware / increases productivity / usage. Also setting -O3 in certain cases instead of -O2 can give some more performance benefits, but it's not recommended as an overall solution, as it can make certain packages actually slower or cause instability. This should be specified on a per package basis. and depending on the usage.

[dimko]

set the CPU Govenor on Performance (desirable on workstation, but undesirable on laptops)

I honestly don't see reason to use Performance governor. Aint it just sets higher power state(with speed benefit) which in turn:
1. disables turboboost on ryzen
2. increases over all temperature(and in dusty desktop this can easily trigger thermo throteling)
3. Consumes more energy without giving much benefit.(imagine, game or some app uses 2-3 cores, but i have 16 core CPU! what a waste!)

[sn99]

disables turboboost on ryzen

I did not knew about this, so we should use ondemand ?