SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2017 INTrate: Trunk generic -O2 - standard vs PGO vs LTO+PGO

-O2

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark Normal PGO LTO+PGO PGO LTO+PGO
500.perlbench_r 100.00% 93.16% 89.62% 100.00% 96.21%
502.gcc_r 100.00% 93.67% 90.41% 100.00% 96.51%
505.mcf_r 100.00% 98.13% 89.46% 100.00% 91.17%
520.omnetpp_r 100.00% 83.94% 78.46% 100.00% 93.46%
523.xalancbmk_r 100.00% 97.04% 91.54% 100.00% 94.34%
525.x264_r 100.00% 71.49% 77.60% 100.00% 108.54%
531.deepsjeng_r 100.00% 109.42% 91.70% 100.00% 83.81%
541.leela_r 100.00% 93.36% 85.71% 100.00% 91.81%
548.exchange2_r 100.00% 64.84% 65.71% 100.00% 101.36%
557.xz_r 100.00% 97.65% 98.08% 100.00% 100.44%
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
500.perlbench_r 100.00% 91.31% 86.26% 100.00% 94.46%
502.gcc_r 100.00% 95.76% 89.95% 100.00% 93.93%
505.mcf_r 100.00% 104.40% 74.97% 100.00% 71.81%
520.omnetpp_r 100.00% 76.08% 63.13% 100.00% 82.98%
523.xalancbmk_r 100.00% 76.87% 59.54% 100.00% 77.46%
525.x264_r 100.00% 108.29% 108.27% 100.00% 99.98%
531.deepsjeng_r 100.00% 100.50% 99.36% 100.00% 98.87%
541.leela_r 100.00% 82.37% 50.30% 100.00% 61.06%
548.exchange2_r 100.00% 278.17% 161.23% 100.00% 57.96%
557.xz_r 100.00% 90.69% 57.70% 100.00% 63.62%
Smaller binary sizes are always better.

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

Run-times

Benchmark PGO LTO+PGO
500.perlbench_r 100.00% 95.00%
502.gcc_r 100.00% 96.60%
505.mcf_r 100.00% 90.00%
520.omnetpp_r 100.00% 92.48%
523.xalancbmk_r 100.00% 93.49%
525.x264_r 100.00% 104.19%
531.deepsjeng_r 100.00% 87.32%
541.leela_r 100.00% 89.44%
548.exchange2_r 100.00% 99.20%
557.xz_r 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 PGO LTO+PGO
500.perlbench_r 100.00% 94.46%
502.gcc_r 100.00% 93.94%
505.mcf_r 100.00% 72.01%
520.omnetpp_r 100.00% 82.98%
523.xalancbmk_r 100.00% 77.46%
525.x264_r 100.00% 99.97%
531.deepsjeng_r 100.00% 98.87%
541.leela_r 100.00% 61.11%
548.exchange2_r 100.00% 57.97%
557.xz_r 100.00% 63.61%
Smaller binary sizes are always better.