SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2017 FPrate: LTO -O2 generic - LLVM and GCC

-O2 -flto

Please note that the same "optimization levels" mean different things to different compilers. For example, LLVM and LLVM-based compiler tend to be much more aggressive at -O2, often with the corresponding consequences to size of the generated binaries and compilation time. GCC is sometimes more aggressive at -O3. LLVM is also often by default less strict when it comes to standard compliance by default and performs optimizations GCC would only do with a special compilation option or at -Ofast. Keep this is mind when comparing different compilers.

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark LLVM GCC
503.bwaves_r 100.00% 104.15%
507.cactuBSSN_r 100.00% 86.47%
508.namd_r 100.00% 110.48%
510.parest_r 100.00% 101.95%
511.povray_r 100.00% 102.12%
519.lbm_r 100.00% 96.23%
521.wrf_r 100.00% 112.09%
526.blender_r 100.00% 102.82%
527.cam4_r 100.00% 98.44%
538.imagick_r 100.00% 121.96%
544.nab_r 100.00% 97.14%
549.fotonik3d_r 100.00% 99.58%
554.roms_r 100.00% 89.39%
LLVM clang version 23.1.3 (/labs/mjambor/git/llvm 0d261d1ca552c95a8f007e061c787ac7132fbcbc), flang version 23.1.3 (/labs/mjambor/git/llvm 0d261d1ca552c95a8f007e061c787ac7132fbcbc)
GCC GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d)
Naturally, the smaller the run-times, the better.

Non-debug executable sizes

Benchmark LLVM GCC
503.bwaves_r 100.00% 4.22%
507.cactuBSSN_r 100.00% 80.71%
508.namd_r 100.00% 102.28%
510.parest_r 100.00% 87.40%
511.povray_r 100.00% 81.72%
519.lbm_r 100.00% 82.31%
521.wrf_r 100.00% 103.71%
526.blender_r 100.00% 75.71%
527.cam4_r 100.00% 103.55%
538.imagick_r 100.00% 90.33%
544.nab_r 100.00% 100.01%
549.fotonik3d_r 100.00% 27.55%
554.roms_r 100.00% 47.89%
Smaller binary sizes are always better.

AMD EPYC 75F3 (zen3 based high-frequency Milan)

Run-times

Benchmark LLVM GCC
503.bwaves_r 100.00% 98.53%
507.cactuBSSN_r 100.00% 85.08%
508.namd_r 100.00% 98.78%
510.parest_r 100.00% 99.53%
511.povray_r 100.00% 100.63%
519.lbm_r 100.00% 94.96%
521.wrf_r 100.00% 116.43%
526.blender_r 100.00% 106.17%
527.cam4_r 100.00% 98.37%
538.imagick_r 100.00% 125.85%
544.nab_r 100.00% 101.53%
549.fotonik3d_r 100.00% 93.66%
554.roms_r 100.00% 92.90%
LLVM clang version 23.1.3 (https://github.com/llvm/llvm-project.git 0d261d1ca552c95a8f007e061c787ac7132fbcbc), flang version 23.1.3 (https://github.com/llvm/llvm-project.git 0d261d1ca552c95a8f007e061c787ac7132fbcbc)
GCC GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d)
Naturally, the smaller the run-times, the better.

Non-debug executable sizes

Benchmark LLVM GCC
503.bwaves_r 100.00% 4.23%
507.cactuBSSN_r 100.00% 80.71%
508.namd_r 100.00% 102.29%
510.parest_r 100.00% 88.00%
511.povray_r 100.00% 81.72%
519.lbm_r 100.00% 82.60%
521.wrf_r 100.00% 103.71%
526.blender_r 100.00% 75.71%
527.cam4_r 100.00% 103.55%
538.imagick_r 100.00% 90.33%
544.nab_r 100.00% 100.01%
549.fotonik3d_r 100.00% 27.56%
554.roms_r 100.00% 47.89%
Smaller binary sizes are always better.