SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2017 INTrate: Trunk LTO -O3 - generic vs x86-64-v3 vs native

-O3 -flto vs -O3 -flto and a given -march

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark Generic x86-64-v3 Native x86-64-v3 Native
500.perlbench_r 100.00% 102.70% 102.70% 100.00% 100.00%
502.gcc_r 100.00% 100.87% 100.87% 100.00% 100.00%
505.mcf_r 100.00% 100.55% 101.10% 100.00% 100.55%
520.omnetpp_r 100.00% 102.31% 100.77% 100.00% 98.50%
523.xalancbmk_r 100.00% 94.94% 93.01% 100.00% 97.97%
525.x264_r 100.00% 88.73% 88.73% 100.00% 100.00%
531.deepsjeng_r 100.00% 94.79% 94.79% 100.00% 100.00%
541.leela_r 100.00% 100.97% 102.17% 100.00% 101.20%
548.exchange2_r 100.00% 100.00% 100.00% 100.00% 100.00%
557.xz_r 100.00% 102.75% 100.85% 100.00% 98.14%
The compiler used was GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d).
Naturally, the smaller the run-times, the better.
See also continuous gcc tuning benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/tuning.

Non-debug executable sizes

Benchmark Generic x86-64-v3 Native x86-64-v3 Native
500.perlbench_r 100.00% 101.16% 102.27% 100.00% 101.09%
502.gcc_r 100.00% 102.95% 104.37% 100.00% 101.38%
505.mcf_r 100.00% 105.89% 110.11% 100.00% 103.98%
520.omnetpp_r 100.00% 100.87% 101.99% 100.00% 101.11%
523.xalancbmk_r 100.00% 104.39% 104.81% 100.00% 100.40%
525.x264_r 100.00% 95.64% 94.22% 100.00% 98.51%
531.deepsjeng_r 100.00% 100.04% 100.15% 100.00% 100.11%
541.leela_r 100.00% 100.93% 100.94% 100.00% 100.01%
548.exchange2_r 100.00% 127.25% 121.35% 100.00% 95.37%
557.xz_r 100.00% 101.86% 101.16% 100.00% 99.31%
Smaller binary sizes are always better.

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

Run-times

Benchmark Generic x86-64-v3 Native x86-64-v3 Native
500.perlbench_r 100.00% 101.39% 101.39% 100.00% 100.00%
502.gcc_r 100.00% 100.00% 100.00% 100.00% 100.00%
505.mcf_r 100.00% 101.11% 100.00% 100.00% 98.90%
520.omnetpp_r 100.00% 100.92% 98.85% 100.00% 97.94%
523.xalancbmk_r 100.00% 92.91% 94.09% 100.00% 101.27%
525.x264_r 100.00% 90.52% 89.54% 100.00% 98.92%
531.deepsjeng_r 100.00% 95.15% 95.63% 100.00% 100.51%
541.leela_r 100.00% 100.94% 101.41% 100.00% 100.47%
548.exchange2_r 100.00% 100.90% 101.80% 100.00% 100.89%
557.xz_r 100.00% 102.11% 101.69% 100.00% 99.59%
The compiler used was GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d).
Naturally, the smaller the run-times, the better.
See also continuous gcc tuning benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/tuning.

Non-debug executable sizes

Benchmark Generic x86-64-v3 Native x86-64-v3 Native
500.perlbench_r 100.00% 101.16% 101.99% 100.00% 100.82%
502.gcc_r 100.00% 102.95% 104.11% 100.00% 101.13%
505.mcf_r 100.00% 105.63% 105.14% 100.00% 99.53%
520.omnetpp_r 100.00% 100.87% 101.76% 100.00% 100.88%
523.xalancbmk_r 100.00% 104.39% 105.92% 100.00% 101.47%
525.x264_r 100.00% 95.67% 96.00% 100.00% 100.35%
531.deepsjeng_r 100.00% 100.04% 100.05% 100.00% 100.01%
541.leela_r 100.00% 100.93% 101.25% 100.00% 100.32%
548.exchange2_r 100.00% 127.24% 124.60% 100.00% 97.92%
557.xz_r 100.00% 101.90% 101.93% 100.00% 100.03%
Smaller binary sizes are always better.