Benchmarking OpenMandriva's AMD Ryzen Optimized Linux Distribution On The Threadripper 3970X

[mroche]

I also wonder if Zen 2 systems would see any difference if running with znver2 instead of znver1.

Cheers,
Mike

[berolinux]

I also wonder if Zen 2 systems would see any difference if running with znver2 instead of znver1.

Looking at the patches for znver2 support in clang and gcc, the difference should be minimal. The timings are virtually the same, and the couple of new instructions added in version 2 don't look like something generally useful that compilers would generate while working on anything resembling normal code.
If anything, maybe the improved avx2 support will cause compilers to generate a few more avx2 instructions.

[mroche]

If anything, maybe the improved avx2 support will cause compilers to generate a few more avx2 instructions.

That’s what I was mostly curious about. I would think the true AVX2 support “should” provide some kind of improvement, but I don’t know enough about compilers and optimizations to know.

Cheers,
Mike

[azdaha]

OpenMandriva is almost there -- glibc is one of the last packages that steadfastly refuse to be compiled with anything other than gcc.

Is using something like musl instead of glibc a viable option?

[azdaha]

Interesting benchmarks although I hoped for better numbers for the zenver1 optimized build.

When compared to Manjaro, I'd say it's a huge improvement. I wonder how Arch stacks up there. I hope it's not as bad

[ms178]

When compared to Manjaro, I'd say it's a huge improvement. I wonder how Arch stacks up there. I hope it's not as bad

Manjaro's default Kernel config is not optimized for performance [e.g. they are not using the performance governor and use some higher security settings], I compiled a custom Kernel on Manjaro with the BMQ scheduler patched in, aggressive compiler flags, throwing out the security stuff and customized the config, and I came within 10% of Clear Linux with the same treatment in my usual game benchmark (52 fps vs 56 fps in Company of Heroes 2). Even though Manjaro did let me down before (stability and performance issues), it earned my respect with makepkg and the rather easy way to custom compile my favorite projects (it is not yet perfect though).

[azdaha]

Manjaro's default Kernel config is not optimized for performance [e.g. they are not using the performance governor and use some higher security settings], I compiled a custom Kernel on Manjaro with the BMQ scheduler patched in, aggressive compiler flags, throwing out the security stuff and customized the config, and I came within 10% of Clear Linux with the same treatment in my usual game benchmark (52 fps vs 56 fps in Company of Heroes 2). Even though Manjaro did let me down before (stability and performance issues), it earned my respect with makepkg and the rather easy way to custom compile my favorite projects (it is not yet perfect though).

Thank you for the details. Although, some of those benefits you mention are really Archlinux. I suspect the kernel configs are the same for Manjaro and Arch, since both use a generic kernel package (aside from the zen kernel option). Maybe it's time for me to do a benchmark then...