SPEC CPU result comparison

15th October 2026 - Normalized report

SPECFP 2006: LTO -Ofast native - LLVM and GCC

-Ofast -march=native -flto

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark LLVM GCC
410.bwaves 100.00% 83.06%
416.gamess 100.00% 108.48%
433.milc 100.00% 84.93%
434.zeusmp 100.00% 112.50%
435.gromacs 100.00% 87.99%
436.cactusADM 100.00% 89.73%
437.leslie3d 100.00% 103.36%
444.namd 100.00% 112.50%
447.dealII 100.00% 97.85%
450.soplex 100.00% 103.32%
453.povray RE RE
454.calculix 100.00% 70.46%
459.GemsFDTD 100.00% 102.86%
465.tonto 100.00% 65.04%
470.lbm 100.00% 54.08%
481.wrf 100.00% 64.69%
482.sphinx3 100.00% 101.61%
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
410.bwaves 100.00% 4.30%
416.gamess 100.00% 99.85%
433.milc 100.00% 80.07%
434.zeusmp 100.00% 124.73%
435.gromacs 100.00% 87.69%
436.cactusADM 100.00% 41.01%
437.leslie3d 100.00% 15.64%
444.namd 100.00% 142.47%
447.dealII 100.00% 88.07%
450.soplex 100.00% 90.39%
453.povray RE RE
454.calculix 100.00% 60.89%
459.GemsFDTD 100.00% 31.61%
465.tonto 100.00% 44.34%
470.lbm 100.00% 52.29%
481.wrf 100.00% 84.48%
482.sphinx3 100.00% 86.42%
Smaller binary sizes are always better.

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

Run-times

Benchmark LLVM GCC
410.bwaves 100.00% 84.31%
416.gamess 100.00% 112.24%
433.milc 100.00% 81.63%
434.zeusmp 100.00% 105.78%
435.gromacs 100.00% 83.22%
436.cactusADM 100.00% 90.04%
437.leslie3d 100.00% 109.35%
444.namd 100.00% 111.35%
447.dealII 100.00% 97.19%
450.soplex 100.00% 103.83%
453.povray RE RE
454.calculix 100.00% 56.82%
459.GemsFDTD 100.00% 100.88%
465.tonto 100.00% 71.47%
470.lbm 100.00% 94.27%
481.wrf 100.00% 63.97%
482.sphinx3 100.00% 99.48%
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
410.bwaves 100.00% 4.37%
416.gamess 100.00% 99.97%
433.milc 100.00% 84.54%
434.zeusmp 100.00% 124.73%
435.gromacs 100.00% 89.85%
436.cactusADM 100.00% 43.22%
437.leslie3d 100.00% 16.28%
444.namd 100.00% 144.86%
447.dealII 100.00% 92.98%
450.soplex 100.00% 92.69%
453.povray RE RE
454.calculix 100.00% 61.71%
459.GemsFDTD 100.00% 32.52%
465.tonto 100.00% 47.23%
470.lbm 100.00% 72.96%
481.wrf 100.00% 90.01%
482.sphinx3 100.00% 87.41%
Smaller binary sizes are always better.