SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2026 FPrate: Trunk LTO -Ofast native - using 128/256/512bit vectors vs default

-Ofast -march=native -mtune=native with and without -mprefer-vector-width=128,256 and/or 512

Please note that event though the header of this comparison says -Ofast, benchmark 767.nest_r was additionally compiled with -fno-finite-math-only. Furthermore, benchmarks 737.gmsh_r and 748.flightdm_r were compiled with -fno-fast-math, so a big chunk of -Ofast was in fact disabled. (This might put GCC at a disadvantage compared to LLVM-based compilers because the latter defaults to -fno-math-errno and -fno-trapping-math, in future measurements, we will try to enable these flags.)

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark 128 256 Default 256 Default
709.cactus_r 100.00% 107.17% 108.30% 100.00% 101.06%
722.palm_r 100.00% 93.32% 93.32% 100.00% 100.00%
731.astcenc_r 100.00% 102.32% 101.66% 100.00% 99.35%
736.ocio_r 100.00% 100.84% 97.20% 100.00% 96.39%
737.gmsh_r 100.00% 100.92% 101.83% 100.00% 100.91%
748.flightdm_r 100.00% 99.01% 101.99% 100.00% 103.01%
749.fotonik3d_r 100.00% 99.52% 92.74% 100.00% 93.19%
765.roms_r 100.00% 82.65% 75.17% 100.00% 90.95%
766.femflow_r 100.00% 59.55% 52.81% 100.00% 88.68%
767.nest_r 100.00% 101.99% 101.99% 100.00% 100.00%
772.marian_r 100.00% 81.82% 81.82% 100.00% 100.00%
782.lbm_r 100.00% 102.01% 103.52% 100.00% 101.48%
The compiler used was GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d).
Naturally, the smaller the run-times, the better.
See also continuous gcc tuning benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/tuning.

Non-debug executable sizes

Benchmark 128 256 Default 256 Default
709.cactus_r 100.00% 102.84% 104.56% 100.00% 101.67%
722.palm_r 100.00% 102.72% 104.25% 100.00% 101.48%
731.astcenc_r 100.00% 116.58% 121.16% 100.00% 103.93%
736.ocio_r 100.00% 101.10% 102.64% 100.00% 101.52%
737.gmsh_r 100.00% 101.55% 103.65% 100.00% 102.06%
748.flightdm_r 100.00% 99.61% 99.79% 100.00% 100.19%
749.fotonik3d_r 100.00% 118.05% 125.70% 100.00% 106.48%
765.roms_r 100.00% 103.84% 105.53% 100.00% 101.63%
766.femflow_r 100.00% 98.69% 100.59% 100.00% 101.93%
767.nest_r 100.00% 100.06% 100.68% 100.00% 100.62%
772.marian_r 100.00% 100.56% 101.34% 100.00% 100.77%
782.lbm_r 100.00% 101.45% 102.54% 100.00% 101.07%
Smaller binary sizes are always better.

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

Run-times

Benchmark 128 Default
709.cactus_r 100.00% 100.86%
722.palm_r 100.00% 92.51%
731.astcenc_r 100.00% 103.35%
736.ocio_r 100.00% 100.84%
737.gmsh_r 100.00% 100.00%
748.flightdm_r 100.00% 99.51%
749.fotonik3d_r 100.00% 93.30%
765.roms_r 100.00% 84.92%
766.femflow_r 100.00% 63.56%
767.nest_r 100.00% 101.33%
772.marian_r 100.00% 103.57%
782.lbm_r 100.00% 99.18%
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 128 Default
709.cactus_r 100.00% 104.15%
722.palm_r 100.00% 104.04%
731.astcenc_r 100.00% 108.93%
736.ocio_r 100.00% 101.38%
737.gmsh_r 100.00% 102.00%
748.flightdm_r 100.00% 99.63%
749.fotonik3d_r 100.00% 122.66%
765.roms_r 100.00% 106.41%
766.femflow_r 100.00% 99.25%
767.nest_r 100.00% 99.64%
772.marian_r 100.00% 101.17%
782.lbm_r 100.00% 101.14%
Smaller binary sizes are always better.