SPEC CPU result comparison

15th October 2026 - Normalized report

SPECINT 2006: PGO+LTO -Ofast native - LLVM, ICX and GCC

-Ofast -native -flto -fprofile-instr-use vs -Ofast -march=native -flto -fprofile-use

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

Run-times

Benchmark LLVM ICX GCC ICX GCC
400.perlbench 100.00% 96.82% 81.74% 100.00% 84.43%
401.bzip2 100.00% 99.79% 96.66% 100.00% 96.86%
403.gcc 100.00% 93.79% 89.58% 100.00% 95.51%
429.mcf 100.00% 92.86% 94.29% 100.00% 101.54%
445.gobmk FE FE FE
456.hmmer 100.00% 85.51% 43.00% 100.00% 50.28%
458.sjeng 100.00% 107.39% 106.30% 100.00% 98.99%
462.libquantum 100.00% 79.79% 88.04% 100.00% 110.34%
464.h264ref 100.00% 96.17% 95.72% 100.00% 99.53%
471.omnetpp 100.00% 97.26% 90.95% 100.00% 93.51%
473.astar 100.00% 89.94% FE 100.00% FE
483.xalancbmk 100.00% 94.55% 84.04% 100.00% 88.89%
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)
ICX Intel(R) oneAPI DPC++/C++ Compiler 2026.1.1 (2026.1.1.20260724), ifx (IFX) 2026.1.1 20260724
GCC GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d)
Naturally, the smaller the run-times, the better.

Non-debug executable sizes

Benchmark LLVM ICX GCC ICX GCC
400.perlbench 100.00% 123.97% 104.72% 100.00% 84.48%
401.bzip2 100.00% 177.23% 68.40% 100.00% 38.59%
403.gcc 100.00% 109.04% 89.38% 100.00% 81.98%
429.mcf 100.00% 189.14% 72.16% 100.00% 38.15%
445.gobmk FE FE FE
456.hmmer 100.00% 184.92% 88.31% 100.00% 47.76%
458.sjeng 100.00% 104.26% 99.34% 100.00% 95.28%
462.libquantum 100.00% 565.97% 75.17% 100.00% 13.28%
464.h264ref 100.00% 136.64% 71.70% 100.00% 52.47%
471.omnetpp 100.00% 140.89% 85.66% 100.00% 60.80%
473.astar 100.00% 278.99% FE 100.00% FE
483.xalancbmk 100.00% 102.97% 83.82% 100.00% 81.40%
Smaller binary sizes are always better.