SPEC CPU result comparison

15th October 2026 - Normalized report

SPECFP 2006: Trunk native -Ofast - standard vs LTO vs LTO+PGO

-Ofast -march=native -mtune=native

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark LTO LTO+PGO
410.bwaves 100.00% 100.00%
416.gamess 100.00% 95.79%
433.milc 100.00% 101.21%
434.zeusmp 100.00% 96.53%
435.gromacs 100.00% 98.01%
436.cactusADM 100.00% 100.00%
437.leslie3d 100.00% 104.55%
444.namd 100.00% 100.68%
447.dealII 100.00% 101.10%
450.soplex 100.00% 96.79%
453.povray RE FE
454.calculix 100.00% 212.12%
459.GemsFDTD 100.00% 98.23%
465.tonto 100.00% 93.39%
470.lbm 100.00% 109.55%
481.wrf 100.00% 108.72%
482.sphinx3 100.00% 95.56%
The compiler used was GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d).
Naturally, the smaller the run-times, the better.
See also continuous gcc options benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/options.

Non-debug executable sizes

Benchmark LTO LTO+PGO
410.bwaves 100.00% 94.09%
416.gamess 100.00% 99.22%
433.milc 100.00% 85.70%
434.zeusmp 100.00% 99.98%
435.gromacs 100.00% 69.76%
436.cactusADM 100.00% 62.78%
437.leslie3d 100.00% 67.86%
444.namd 100.00% 54.66%
447.dealII 100.00% 64.66%
450.soplex 100.00% 64.70%
453.povray RE FE
454.calculix 100.00% 66.18%
459.GemsFDTD 100.00% 68.56%
465.tonto 100.00% 66.65%
470.lbm 100.00% 101.45%
481.wrf 100.00% 72.29%
482.sphinx3 100.00% 77.09%
Smaller binary sizes are always better.

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

Run-times

Benchmark LTO LTO+PGO
410.bwaves 100.00% 101.26%
416.gamess 100.00% 97.53%
433.milc 100.00% 110.82%
434.zeusmp 100.00% 98.56%
435.gromacs 100.00% 100.80%
436.cactusADM 100.00% 101.77%
437.leslie3d 100.00% 100.00%
444.namd 100.00% 96.75%
447.dealII 100.00% 100.00%
450.soplex 100.00% 98.16%
453.povray RE FE
454.calculix 100.00% 222.86%
459.GemsFDTD 100.00% 101.30%
465.tonto 100.00% 95.32%
470.lbm 100.00% 91.22%
481.wrf 100.00% 111.08%
482.sphinx3 100.00% 99.48%
The compiler used was GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d).
Naturally, the smaller the run-times, the better.
See also continuous gcc options benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/options.

Non-debug executable sizes

Benchmark LTO LTO+PGO
410.bwaves 100.00% 95.55%
416.gamess 100.00% 99.23%
433.milc 100.00% 87.01%
434.zeusmp 100.00% 99.98%
435.gromacs 100.00% 70.28%
436.cactusADM 100.00% 61.59%
437.leslie3d 100.00% 67.48%
444.namd 100.00% 52.49%
447.dealII 100.00% 65.30%
450.soplex 100.00% 62.62%
453.povray RE FE
454.calculix 100.00% 66.51%
459.GemsFDTD 100.00% 67.09%
465.tonto 100.00% 65.09%
470.lbm 100.00% 100.66%
481.wrf 100.00% 70.82%
482.sphinx3 100.00% 75.33%
Smaller binary sizes are always better.