SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2026 FPrate: Trunk native -Ofast - standard vs PGO vs LTO+PGO

-Ofast -march=native -mtune=native

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 Normal PGO LTO+PGO PGO LTO+PGO
709.cactus_r 100.00% 97.36% 104.53% 100.00% 107.36%
722.palm_r 100.00% 97.42% 96.85% 100.00% 99.41%
731.astcenc_r 100.00% 118.42% 114.80% 100.00% 96.94%
736.ocio_r 100.00% 100.00% 99.14% 100.00% 99.14%
737.gmsh_r 100.00% 100.00% 98.23% 100.00% 98.23%
748.flightdm_r 100.00% 92.84% 80.54% 100.00% 86.75%
749.fotonik3d_r 100.00% 111.38% 112.70% 100.00% 101.19%
765.roms_r 100.00% 100.00% 101.35% 100.00% 101.35%
766.femflow_r 100.00% CE 102.17% CE
767.nest_r 100.00% 107.98% 103.13% 100.00% 95.51%
772.marian_r 100.00% 111.11% 122.22% 100.00% 110.00%
782.lbm_r 100.00% 96.63% 96.63% 100.00% 100.00%
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 Normal PGO LTO+PGO PGO LTO+PGO
709.cactus_r 100.00% 73.28% 61.15% 100.00% 83.45%
722.palm_r 100.00% 80.30% 71.28% 100.00% 88.77%
731.astcenc_r 100.00% 59.32% 49.62% 100.00% 83.65%
736.ocio_r 100.00% 48.67% 33.25% 100.00% 68.32%
737.gmsh_r 100.00% 58.41% 36.24% 100.00% 62.05%
748.flightdm_r 100.00% 68.56% 61.07% 100.00% 89.07%
749.fotonik3d_r 100.00% 65.84% 60.10% 100.00% 91.28%
765.roms_r 100.00% 84.57% 68.29% 100.00% 80.75%
766.femflow_r 100.00% CE 3.80% CE
767.nest_r 100.00% 61.85% 50.82% 100.00% 82.16%
772.marian_r 100.00% 68.36% 47.55% 100.00% 69.56%
782.lbm_r 100.00% 90.82% 74.10% 100.00% 81.59%
Smaller binary sizes are always better.

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

Run-times

Benchmark Normal PGO LTO+PGO PGO LTO+PGO
709.cactus_r 100.00% 99.43% 99.43% 100.00% 100.00%
722.palm_r 100.00% 95.58% 96.10% 100.00% 100.54%
731.astcenc_r 100.00% 114.37% 109.09% 100.00% 95.38%
736.ocio_r 100.00% 100.00% 99.45% 100.00% 99.45%
737.gmsh_r 100.00% 101.30% 97.17% 100.00% 95.92%
748.flightdm_r 100.00% 92.79% 79.26% 100.00% 85.41%
749.fotonik3d_r 100.00% 102.84% 109.04% 100.00% 106.03%
765.roms_r 100.00% 100.72% 100.72% 100.00% 100.00%
766.femflow_r 100.00% 194.52% 104.11% 100.00% 53.52%
767.nest_r 100.00% 107.75% 102.58% 100.00% 95.20%
772.marian_r 100.00% 103.45% 106.90% 100.00% 103.33%
782.lbm_r 100.00% 101.45% 100.00% 100.00% 98.57%
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 Normal PGO LTO+PGO PGO LTO+PGO
709.cactus_r 100.00% 73.73% 61.72% 100.00% 83.71%
722.palm_r 100.00% 80.34% 71.35% 100.00% 88.81%
731.astcenc_r 100.00% 57.81% 46.33% 100.00% 80.15%
736.ocio_r 100.00% 48.76% 33.12% 100.00% 67.93%
737.gmsh_r 100.00% 59.13% 36.59% 100.00% 61.87%
748.flightdm_r 100.00% 68.76% 61.14% 100.00% 88.91%
749.fotonik3d_r 100.00% 63.68% 58.38% 100.00% 91.67%
765.roms_r 100.00% 84.20% 68.11% 100.00% 80.89%
766.femflow_r 100.00% 45.32% 3.92% 100.00% 8.66%
767.nest_r 100.00% 62.18% 51.15% 100.00% 82.26%
772.marian_r 100.00% 67.75% 47.72% 100.00% 70.44%
782.lbm_r 100.00% 88.71% 74.47% 100.00% 83.94%
Smaller binary sizes are always better.