Initial Benchmarks Of The AMD AOCC 5.0 Compiler On 5th Gen EPYC

[phoronix]

Phoronix: Initial Benchmarks Of The AMD AOCC 5.0 Compiler On 5th Gen EPYC

Last week when launching the AMD EPYC 9005 "Turin" processors, on the same day AOCC 5.0 was quietly released as the newest version of AMD's Zen-focused compiler derived from LLVM/Clang. With not only adding AMD Zen 5 "znver5" support but also additional vectorization improvements and other performance optimizations, I was eager to run some benchmarks of AOCC 5.0 against the open-source GCC and LLVM/Clang compilers. Here are those initial benchmarks using dual AMD EPYC 9755 128-core Zen 5 processors.

https://www.phoronix.com/review/amd-aocc-5

[sophisticles]

I'm of the school of thought that says that you should always use the vendor supplied compiler, even when there is little immediately noticeable difference.

To me AMD went through the trouble of creating this compiler for these processors, use them all the time, in some cases you will see a nice performance gain, in some no gain, but it doesn't look like it hurts performance at all and for the sake of simplicity just use AOCC and call it a day.

[Kjell]

Are there any downsides to using AOCC for everything (apart from slightly outdated Clang)?

[elfurbe]

So I just pulled down aocc 5.0.0 to rebuild SVT-AV1 and I'm seeing a lot of this message during compilation:

clang: warning: Unsupported lto mode -flto=thin, falling back to -flto=full.

I did not get that with aocc 4.2.0.

phoronix did you see this also when you built it SVT-AV1? Maybe AMD forgot to enable thin lto when they built this new clang?

[dkokron]

Does anyone else think including the Intel compilers in this type of comparison would be a tremendous service to the community?

[sabian2008]

Nice benchmarks

Would be great to see some Fortran tests and a comparison with ifort and ifx. I still get my fastest binaries for Fortran HPC code using ifort, even on Epyc nodes.

[lejeczek]

It is not looking bad but neither it looks great - but keep them coming, those test results. Thanks!

[pokeballs]

I'm of the school of thought that says that you should always use the vendor supplied compiler, even when there is little immediately noticeable difference.

To me AMD went through the trouble of creating this compiler for these processors, use them all the time, in some cases you will see a nice performance gain, in some no gain, but it doesn't look like it hurts performance at all and for the sake of simplicity just use AOCC and call it a day.

Intel upstreams directly to GCC and LLVM, why would I use an outdated custom compiler just for AMD ???

[ms178]

It is funny that both AMD and Intel are actually using LLVM/Clang for some time now and add some sugar to their own branded compiler. I wonder if Microsoft will do the same sooner or later with MSVC.

[Azpegath]

I'm pleasantly surprised by the results, my impression from earlier tests was that the difference was minuscule if at all better. But in this case, it was the leader in every single test, though by little is some.

I have a 3950X, I guess that the improvements won't be that big for me?