AOCC 4.0 Shows The Strong Advantages Of Compiler Optimizations With 4th Gen AMD EPYC CPUs
Tencent's TNN neural network software had some rare wins for GCC.
From texture compression to HPC workloads, AOCC 4.0 was quite consistently delivering binaries that were outperforming GCC 12 stable.
In the 68 benchmarks carried out for this round of benchmarking, the AOCC 4.0 built binaries were on average 17% faster than the GCC 12.2 built binaries using the geometric mean for all of the (C/C++) tests carried out. Those interested can see all of the benchmarks in full for closer examination via this result page. I absolutely love the performance uplift with AOCC 4.0 but at the same time wished upstream GCC and LLVM/Clang had their Zen 4 optimizations too ahead of launch rather than what will be months later until being found in released compiler versions.
Given the earlier AOCC 4.0 benchmarks with the Ryzen 9 7950X desktop processor and the lack of tuned upstream support yet for Zen 4, these numbers shouldn't be too surprising but nevertheless important for those wanting to maximize their investment with 4th Gen EPYC or Ryzen 7000 series performance. Hopefully these numbers help you to quantify the possible performance benefits to enjoy with 4th Gen EPYC if wanting to use the AMD Optimizing C/C++ Compiler for squeezing out even greater performance than the stock system compiler. I'll be back around with more Genoa compiler benchmarks once the pending GCC 13 Znver4 material is squared away and seeing what LLVM/Clang patches may be upstreamed in the near-term for further maximizing the performance potential of Zen 4 on the desktop and servers.
If you enjoyed this article consider joining Phoronix Premium to view this site ad-free, multi-page articles on a single page, and other benefits. PayPal or Stripe tips are also graciously accepted. Thanks for your support.