Clang 21 Delivering Nice Performance Gains On AMD EPYC Zen 4 With HBM3
Written by Michael Larabel in Software on 21 November 2025 at 11:00 AM EST. Page 2 of 4. 5 Comments.
👁 NAS Parallel Benchmarks benchmark with settings of Test / Class: BT.C. Clang 21.1 was the fastest.
👁 NAS Parallel Benchmarks benchmark with settings of Test / Class: EP.C. Clang 21.1 was the fastest.
👁 NAS Parallel Benchmarks benchmark with settings of Test / Class: LU.C. Clang 20.1 was the fastest.
👁 NAS Parallel Benchmarks benchmark with settings of Test / Class: SP.B. Clang 21.1 was the fastest.
👁 NAS Parallel Benchmarks benchmark with settings of Test / Class: SP.C. Clang 21.1 was the fastest.
Sure enough, even with Zen 4 there were very nice gains still being observed from the LLVM/Clang compiler upgrade. Zen 4 with "-march=native" optimized builds was still showing some nice incremental performance gains on LLVM/Clang 21 even with being a more mature generation compared to the prior Zen 5 / Turin benchmarks.
The Azure HBv5 instance was already a mighty performer thanks to the EPYC 9V64H CPU with HBM3 memory while the LLVM/Clang compiler upgrade can help tap some additional performance gains for high performance computing.