SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2017 FPrate: PGO+LTO -Ofast native - LLVM, AOCC and GCC

-Ofast -march=native -mtune=native -flto and -fprofile-instr-use vs. -profile-use

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark LLVM AOCC GCC AOCC GCC
503.bwaves_r CE 100.00% 87.96%
507.cactuBSSN_r CE 100.00% 118.10%
508.namd_r 100.00% 101.11% 111.36% 100.00% 110.14%
510.parest_r 100.00% 95.64% 85.45% 100.00% 89.35%
511.povray_r 100.00% 111.33% 88.40% 100.00% 79.40%
519.lbm_r 100.00% 51.15% 48.62% 100.00% 95.05%
521.wrf_r CE 100.00% 93.33%
526.blender_r 100.00% 73.29% 102.84% 100.00% 140.32%
527.cam4_r CE 100.00% 83.71%
538.imagick_r 100.00% 86.16% 175.47% 100.00% 203.65%
544.nab_r 100.00% 110.63% 101.72% 100.00% 91.95%
549.fotonik3d_r CE 100.00% 103.56%
554.roms_r CE 100.00% 96.71%
LLVM clang version 23.1.3 (/labs/mjambor/git/llvm 0d261d1ca552c95a8f007e061c787ac7132fbcbc), flang version 23.1.3 (/labs/mjambor/git/llvm 0d261d1ca552c95a8f007e061c787ac7132fbcbc)
AOCC AMD clang version 17.0.6 (CLANG: AOCC_5.2.0-Build#2035 2026_04_10)
GCC GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d)
In the case of LLVM -fno-finite-math-only was used to compile 511.povray_r (that benchmark only) to avoid a known issue.
Naturally, the smaller the run-times, the better.

Non-debug executable sizes

Benchmark LLVM AOCC GCC AOCC GCC
503.bwaves_r CE 100.00% 65.06%
507.cactuBSSN_r CE 100.00% 50.92%
508.namd_r 100.00% 99.74% 84.54% 100.00% 84.76%
510.parest_r 100.00% 100.30% 71.99% 100.00% 71.78%
511.povray_r 100.00% 106.24% 82.25% 100.00% 77.42%
519.lbm_r 100.00% 125.07% 59.02% 100.00% 47.19%
521.wrf_r CE 100.00% 73.05%
526.blender_r 100.00% 104.73% 68.18% 100.00% 65.10%
527.cam4_r CE 100.00% 81.38%
538.imagick_r 100.00% 145.27% 77.55% 100.00% 53.38%
544.nab_r 100.00% 105.64% 96.42% 100.00% 91.27%
549.fotonik3d_r CE 100.00% 63.18%
554.roms_r CE 100.00% 72.04%
Smaller binary sizes are always better.