SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2026 FPrate: LTO x86-64-v3 -O3 - Released GCCs vs Trunk

-O3 -flto -march=x86-64-v3

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark GCC 15 GCC 16 Trunk GCC 16 Trunk
709.cactus_r 100.00% 127.69% 129.23% 100.00% 101.20%
722.palm_r 100.00% 101.62% 101.16% 100.00% 99.54%
731.astcenc_r 100.00% 99.05% 95.72% 100.00% 96.64%
736.ocio_r 100.00% 104.57% 104.57% 100.00% 100.00%
737.gmsh_r 100.00% 100.00% 100.00% 100.00% 100.00%
748.flightdm_r 100.00% 100.00% 98.03% 100.00% 98.03%
749.fotonik3d_r 100.00% 99.28% 98.80% 100.00% 99.52%
765.roms_r 100.00% 106.93% 99.27% 100.00% 92.83%
766.femflow_r 100.00% 76.00% 80.00% 100.00% 105.26%
767.nest_r 100.00% 120.33% 124.12% 100.00% 103.15%
772.marian_r 100.00% 100.00% 95.00% 100.00% 95.00%
782.lbm_r 100.00% 49.11% 51.56% 100.00% 105.00%
GCC 15 GCC 15.3.0 (revision 4db0e8df15bef836558857c291c323add11d035c)
GCC 16 GCC 16.2.0 (revision 78d4ac73dd391005b895a6148cd9831e28e1208b)
Trunk GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d)
Naturally, the smaller the run-times, the better.
See also continuous gcc branch benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/branch.

Non-debug executable sizes

Benchmark GCC 15 GCC 16 Trunk GCC 16 Trunk
709.cactus_r 100.00% 103.77% 101.35% 100.00% 97.67%
722.palm_r 100.00% 101.74% 100.45% 100.00% 98.73%
731.astcenc_r 100.00% 99.37% 87.00% 100.00% 87.55%
736.ocio_r 100.00% 102.51% 99.91% 100.00% 97.46%
737.gmsh_r 100.00% 103.82% 101.77% 100.00% 98.03%
748.flightdm_r 100.00% 104.55% 99.18% 100.00% 94.87%
749.fotonik3d_r 100.00% 102.47% 99.68% 100.00% 97.27%
765.roms_r 100.00% 102.04% 99.72% 100.00% 97.73%
766.femflow_r 100.00% 101.49% 100.28% 100.00% 98.81%
767.nest_r 100.00% 102.82% 100.70% 100.00% 97.94%
772.marian_r 100.00% 103.56% 101.61% 100.00% 98.12%
782.lbm_r 100.00% 100.00% 97.67% 100.00% 97.67%
Smaller binary sizes are always better.

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

Run-times

Benchmark GCC 15 GCC 16 Trunk GCC 16 Trunk
709.cactus_r 100.00% 120.76% 126.61% 100.00% 104.84%
722.palm_r 100.00% 102.23% 100.45% 100.00% 98.25%
731.astcenc_r 100.00% 98.10% 96.91% 100.00% 98.79%
736.ocio_r 100.00% 99.20% 99.20% 100.00% 100.00%
737.gmsh_r 100.00% 100.00% 98.68% 100.00% 98.68%
748.flightdm_r 100.00% 97.62% 98.10% 100.00% 100.49%
749.fotonik3d_r 100.00% 98.99% 96.73% 100.00% 97.72%
765.roms_r 100.00% 105.88% 98.37% 100.00% 92.90%
766.femflow_r 100.00% 79.35% 85.87% 100.00% 108.22%
767.nest_r 100.00% 117.99% 119.84% 100.00% 101.57%
772.marian_r 100.00% 97.14% 94.29% 100.00% 97.06%
782.lbm_r 100.00% 98.88% 99.33% 100.00% 100.45%
GCC 15 GCC 15.3.0 (revision 4db0e8df15bef836558857c291c323add11d035c)
GCC 16 GCC 16.2.0 (revision 78d4ac73dd391005b895a6148cd9831e28e1208b)
Trunk GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d)
Naturally, the smaller the run-times, the better.
See also continuous gcc branch benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/branch.

Non-debug executable sizes

Benchmark GCC 15 GCC 16 Trunk GCC 16 Trunk
709.cactus_r 100.00% 103.77% 101.35% 100.00% 97.67%
722.palm_r 100.00% 101.74% 100.45% 100.00% 98.73%
731.astcenc_r 100.00% 99.37% 87.00% 100.00% 87.55%
736.ocio_r 100.00% 102.51% 99.91% 100.00% 97.46%
737.gmsh_r 100.00% 103.82% 101.77% 100.00% 98.03%
748.flightdm_r 100.00% 104.55% 99.18% 100.00% 94.87%
749.fotonik3d_r 100.00% 102.47% 99.68% 100.00% 97.27%
765.roms_r 100.00% 102.03% 99.72% 100.00% 97.73%
766.femflow_r 100.00% 101.48% 100.28% 100.00% 98.81%
767.nest_r 100.00% 102.82% 100.70% 100.00% 97.94%
772.marian_r 100.00% 103.56% 101.61% 100.00% 98.12%
782.lbm_r 100.00% 100.46% 98.14% 100.00% 97.68%
Smaller binary sizes are always better.