SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2026 FPrate: Trunk LTO -O2 - generic vs x86-64-v3 vs native

-O2 -flto vs -O2 -flto and a given -march

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark Generic x86-64-v3 Native x86-64-v3 Native
709.cactus_r 100.00% 83.37% 82.88% 100.00% 99.40%
722.palm_r 100.00% 97.15% 91.01% 100.00% 93.68%
731.astcenc_r 100.00% 87.89% 84.60% 100.00% 96.26%
736.ocio_r 100.00% 78.57% 80.00% 100.00% 101.82%
737.gmsh_r 100.00% VE 100.00% VE
748.flightdm_r 100.00% 100.00% 99.11% 100.00% 99.11%
749.fotonik3d_r 100.00% 91.88% 88.17% 100.00% 95.96%
765.roms_r 100.00% 93.23% 90.48% 100.00% 97.04%
766.femflow_r 100.00% 18.02% 15.38% 100.00% 85.37%
767.nest_r 100.00% 100.00% 94.59% 100.00% 94.59%
772.marian_r 100.00% 97.23% 97.23% 100.00% 100.00%
782.lbm_r 100.00% 101.76% 95.59% 100.00% 93.94%
The compiler used was GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d).
Naturally, the smaller the run-times, the better.
See also continuous gcc tuning benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/tuning.

Non-debug executable sizes

Benchmark Generic x86-64-v3 Native x86-64-v3 Native
709.cactus_r 100.00% 93.21% 94.82% 100.00% 101.73%
722.palm_r 100.00% 100.14% 100.85% 100.00% 100.71%
731.astcenc_r 100.00% 91.79% 99.03% 100.00% 107.88%
736.ocio_r 100.00% 100.55% 101.94% 100.00% 101.39%
737.gmsh_r 100.00% VE 100.47% VE
748.flightdm_r 100.00% 99.82% 101.18% 100.00% 101.36%
749.fotonik3d_r 100.00% 100.24% 102.58% 100.00% 102.34%
765.roms_r 100.00% 99.74% 99.02% 100.00% 99.27%
766.femflow_r 100.00% 84.80% 83.34% 100.00% 98.27%
767.nest_r 100.00% 99.71% 100.69% 100.00% 100.98%
772.marian_r 100.00% 101.35% 103.25% 100.00% 101.87%
782.lbm_r 100.00% 95.45% 94.81% 100.00% 99.33%
Smaller binary sizes are always better.

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

Run-times

Benchmark Generic x86-64-v3
709.cactus_r 100.00% 93.12%
722.palm_r 100.00% 96.48%
731.astcenc_r 100.00% 85.31%
736.ocio_r 100.00% 81.73%
737.gmsh_r 100.00% VE
748.flightdm_r 100.00% 99.55%
749.fotonik3d_r 100.00% 86.26%
765.roms_r 100.00% 90.95%
766.femflow_r 100.00% 23.57%
767.nest_r 100.00% 101.52%
772.marian_r 100.00% 98.37%
782.lbm_r 100.00% 98.89%
The compiler used was GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d).
Naturally, the smaller the run-times, the better.
See also continuous gcc tuning benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/tuning.

Non-debug executable sizes

Benchmark Generic x86-64-v3
709.cactus_r 100.00% 93.21%
722.palm_r 100.00% 100.14%
731.astcenc_r 100.00% 91.78%
736.ocio_r 100.00% 100.55%
737.gmsh_r 100.00% VE
748.flightdm_r 100.00% 99.82%
749.fotonik3d_r 100.00% 100.24%
765.roms_r 100.00% 99.74%
766.femflow_r 100.00% 84.80%
767.nest_r 100.00% 99.71%
772.marian_r 100.00% 101.35%
782.lbm_r 100.00% 95.70%
Smaller binary sizes are always better.