SPEC CPU result comparison

15th October 2026 - Normalized report

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

-Ofast -march=native -flto=thin vs. -Ofast -march=native -flto

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 ThinLTO LLVM FullLTO GCC LLVM FullLTO GCC
709.cactus_r 100.00% 95.77% 110.38% 100.00% 115.26%
722.palm_r 100.00% 96.99% 116.72% 100.00% 120.34%
731.astcenc_r 100.00% 100.00% 117.18% 100.00% 117.18%
736.ocio_r 100.00% 98.93% 93.03% 100.00% 94.04%
737.gmsh_r 100.00% 100.00% 93.28% 100.00% 93.28%
748.flightdm_r 100.00% 100.00% 92.78% 100.00% 92.78%
749.fotonik3d_r 100.00% 99.47% 102.13% 100.00% 102.68%
765.roms_r 100.00% 98.75% 92.08% 100.00% 93.25%
766.femflow_r 100.00% 89.09% 85.45% 100.00% 95.92%
767.nest_r 100.00% 100.83% 98.90% 100.00% 98.08%
772.marian_r 100.00% 93.33% 60.00% 100.00% 64.29%
782.lbm_r 100.00% 100.00% 60.23% 100.00% 60.23%
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 ThinLTO LLVM FullLTO GCC LLVM FullLTO GCC
709.cactus_r 100.00% 97.63% 92.88% 100.00% 95.13%
722.palm_r 100.00% 93.66% 77.37% 100.00% 82.60%
731.astcenc_r 100.00% 97.20% 134.76% 100.00% 138.64%
736.ocio_r 100.00% 71.11% 64.16% 100.00% 90.22%
737.gmsh_r 100.00% 97.81% 94.53% 100.00% 96.65%
748.flightdm_r 100.00% 100.77% 92.15% 100.00% 91.45%
749.fotonik3d_r 100.00% 106.51% 37.68% 100.00% 35.38%
765.roms_r 100.00% 93.09% 65.70% 100.00% 70.58%
766.femflow_r 100.00% 115.06% 77.80% 100.00% 67.62%
767.nest_r 100.00% 95.77% 81.40% 100.00% 84.99%
772.marian_r 100.00% 96.51% 90.33% 100.00% 93.60%
782.lbm_r 100.00% 83.53% 37.69% 100.00% 45.12%
Smaller binary sizes are always better.

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

Run-times

Benchmark LLVM ThinLTO LLVM FullLTO GCC LLVM FullLTO GCC
709.cactus_r 100.00% 101.44% 101.44% 100.00% 100.00%
722.palm_r 100.00% 91.77% 116.77% 100.00% 127.24%
731.astcenc_r 100.00% 97.90% 101.80% 100.00% 103.99%
736.ocio_r 100.00% 98.73% 91.35% 100.00% 92.53%
737.gmsh_r 100.00% 99.37% 94.54% 100.00% 95.14%
748.flightdm_r 100.00% 98.86% 93.61% 100.00% 94.69%
749.fotonik3d_r 100.00% 99.20% 103.45% 100.00% 104.28%
765.roms_r 100.00% 97.70% 90.49% 100.00% 92.62%
766.femflow_r 100.00% 94.12% 110.29% 100.00% 117.19%
767.nest_r 100.00% 100.00% 101.88% 100.00% 101.88%
772.marian_r 100.00% 92.59% 107.41% 100.00% 116.00%
782.lbm_r 100.00% 100.96% 115.65% 100.00% 114.56%
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 ThinLTO LLVM FullLTO GCC LLVM FullLTO GCC
709.cactus_r 100.00% 98.74% 95.14% 100.00% 96.35%
722.palm_r 100.00% 97.24% 81.11% 100.00% 83.41%
731.astcenc_r 100.00% 97.71% 156.75% 100.00% 160.42%
736.ocio_r 100.00% 72.67% 66.76% 100.00% 91.86%
737.gmsh_r 100.00% 97.68% 95.31% 100.00% 97.57%
748.flightdm_r 100.00% 100.04% 91.80% 100.00% 91.77%
749.fotonik3d_r 100.00% 104.87% 37.78% 100.00% 36.03%
765.roms_r 100.00% 94.46% 67.89% 100.00% 71.87%
766.femflow_r 100.00% 116.23% 78.00% 100.00% 67.11%
767.nest_r 100.00% 94.80% 82.99% 100.00% 87.55%
772.marian_r 100.00% 96.38% 89.50% 100.00% 92.87%
782.lbm_r 100.00% 83.26% 58.51% 100.00% 70.28%
Smaller binary sizes are always better.