SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2026 FPrate: PGO+ThinLTO and PGO+LTO native -Ofast - LLVM and GCC

-Ofast -march=native -mtune=native -flto=(thin) and -fprofile-instr-use vs. -profile-use

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 LLVM LLVM FullLTO GCC LLVM FullLTO GCC
709.cactus_r 100.00% 102.03% 112.60% 100.00% 110.36%
722.palm_r CE CE CE
731.astcenc_r 100.00% 100.00% 134.23% 100.00% 134.23%
736.ocio_r 100.00% 100.00% 95.56% 100.00% 95.56%
737.gmsh_r 100.00% 98.54% 92.69% 100.00% 94.07%
748.flightdm_r 100.00% 102.86% 85.92% 100.00% 83.53%
749.fotonik3d_r CE CE CE
765.roms_r CE CE CE
766.femflow_r 100.00% 89.09% 85.45% 100.00% 95.92%
767.nest_r 100.00% 80.36% 73.28% 100.00% 91.18%
772.marian_r 100.00% 100.00% 68.75% 100.00% 68.75%
782.lbm_r 100.00% 99.59% 41.02% 100.00% 41.19%
LLVM clang version 23.1.3 (/labs/mjambor/git/llvm 0d261d1ca552c95a8f007e061c787ac7132fbcbc), flang version 23.1.3 (/labs/mjambor/git/llvm 0d261d1ca552c95a8f007e061c787ac7132fbcbc)
GCC GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d)
Naturally, the smaller the run-times, the better.

Non-debug executable sizes

Benchmark LLVM LLVM FullLTO GCC LLVM FullLTO GCC
709.cactus_r 100.00% 95.18% 86.13% 100.00% 90.49%
722.palm_r CE CE CE
731.astcenc_r 100.00% 105.29% 99.16% 100.00% 94.18%
736.ocio_r 100.00% 75.29% 61.83% 100.00% 82.12%
737.gmsh_r 100.00% 99.67% 74.47% 100.00% 74.71%
748.flightdm_r 100.00% 99.64% 99.75% 100.00% 100.11%
749.fotonik3d_r CE CE CE
765.roms_r CE CE CE
766.femflow_r 100.00% 103.05% 66.60% 100.00% 64.63%
767.nest_r 100.00% 97.96% 71.89% 100.00% 73.39%
772.marian_r 100.00% 94.93% 79.63% 100.00% 83.89%
782.lbm_r 100.00% 102.45% 41.59% 100.00% 40.59%
Smaller binary sizes are always better.

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

Run-times

Benchmark LLVM LLVM FullLTO GCC LLVM FullLTO GCC
709.cactus_r 100.00% 101.13% 99.15% 100.00% 98.04%
722.palm_r CE CE CE
731.astcenc_r 100.00% 97.59% 112.05% 100.00% 114.81%
736.ocio_r 100.00% 100.76% 91.39% 100.00% 90.70%
737.gmsh_r 100.00% 99.37% 93.91% 100.00% 94.50%
748.flightdm_r 100.00% 100.48% 87.47% 100.00% 87.05%
749.fotonik3d_r CE CE CE
765.roms_r CE CE CE
766.femflow_r 100.00% 96.97% 115.15% 100.00% 118.75%
767.nest_r 100.00% 95.87% 96.36% 100.00% 100.51%
772.marian_r 100.00% 103.57% 110.71% 100.00% 106.90%
782.lbm_r 100.00% 100.00% 109.18% 100.00% 109.18%
LLVM clang version 23.1.3 (https://github.com/llvm/llvm-project.git 0d261d1ca552c95a8f007e061c787ac7132fbcbc), flang version 23.1.3 (https://github.com/llvm/llvm-project.git 0d261d1ca552c95a8f007e061c787ac7132fbcbc)
GCC GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d)
Naturally, the smaller the run-times, the better.

Non-debug executable sizes

Benchmark LLVM LLVM FullLTO GCC LLVM FullLTO GCC
709.cactus_r 100.00% 94.04% 87.40% 100.00% 92.94%
722.palm_r CE CE CE
731.astcenc_r 100.00% 105.48% 102.29% 100.00% 96.97%
736.ocio_r 100.00% 74.86% 61.96% 100.00% 82.77%
737.gmsh_r 100.00% 99.53% 75.32% 100.00% 75.68%
748.flightdm_r 100.00% 99.77% 99.90% 100.00% 100.14%
749.fotonik3d_r CE CE CE
765.roms_r CE CE CE
766.femflow_r 100.00% 105.47% 70.41% 100.00% 66.76%
767.nest_r 100.00% 97.34% 72.63% 100.00% 74.61%
772.marian_r 100.00% 94.92% 80.52% 100.00% 84.82%
782.lbm_r 100.00% 95.32% 65.83% 100.00% 69.06%
Smaller binary sizes are always better.