SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2026 INTrate: Trunk generic -O2 - standard vs LTO vs LTO+PGO

-O2

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark Normal LTO LTO+PGO LTO LTO+PGO
706.stockfish_r 100.00% 98.69% 101.31% 100.00% 102.65%
707.ntest_r 100.00% 99.37% 98.75% 100.00% 99.37%
708.sqlite_r 100.00% 100.00% 94.12% 100.00% 94.12%
710.omnetpp_r 100.00% 86.01% 75.00% 100.00% 87.20%
714.cpython_r 100.00% 93.26% 76.30% 100.00% 81.82%
721.gcc_r 100.00% 96.70% 90.31% 100.00% 93.39%
723.llvm_r 100.00% 101.48% 84.18% 100.00% 82.95%
727.cppcheck_r 100.00% 93.38% 88.03% 100.00% 94.28%
729.abc_r 100.00% 100.00% 109.05% 100.00% 109.05%
734.vpr_r 100.00% 89.42% 83.80% 100.00% 93.72%
735.gem5_r 100.00% 94.50% 109.40% 100.00% 115.78%
750.sealcrypto_r 100.00% 108.70% 123.19% 100.00% 113.33%
753.ns3_r 100.00% 90.04% 86.99% 100.00% 96.61%
777.zstd_r 100.00% 100.69% 97.22% 100.00% 96.55%
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 LTO LTO+PGO LTO LTO+PGO
706.stockfish_r 100.00% 89.85% 86.90% 100.00% 96.72%
707.ntest_r 100.00% 68.28% 64.70% 100.00% 94.75%
708.sqlite_r 100.00% 96.54% 82.53% 100.00% 85.49%
710.omnetpp_r 100.00% 86.04% 59.66% 100.00% 69.35%
714.cpython_r 100.00% 75.73% 65.24% 100.00% 86.14%
721.gcc_r 100.00% 102.04% 84.30% 100.00% 82.62%
723.llvm_r 100.00% 70.40% 50.73% 100.00% 72.07%
727.cppcheck_r 100.00% 82.25% 59.84% 100.00% 72.76%
729.abc_r 100.00% 69.38% 49.28% 100.00% 71.02%
734.vpr_r 100.00% 78.66% 59.55% 100.00% 75.71%
735.gem5_r 100.00% 71.49% 50.85% 100.00% 71.13%
750.sealcrypto_r 100.00% 45.23% 33.18% 100.00% 73.37%
753.ns3_r 100.00% 73.71% 52.22% 100.00% 70.85%
777.zstd_r 100.00% 84.50% 67.78% 100.00% 80.22%
Smaller binary sizes are always better.

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

Run-times

Benchmark LTO LTO+PGO
706.stockfish_r 100.00% 100.00%
707.ntest_r 100.00% 96.90%
708.sqlite_r 100.00% 92.44%
710.omnetpp_r 100.00% 88.30%
714.cpython_r 100.00% 84.01%
721.gcc_r 100.00% 90.79%
723.llvm_r 100.00% 77.02%
727.cppcheck_r 100.00% 93.39%
729.abc_r 100.00% 110.62%
734.vpr_r 100.00% 94.42%
735.gem5_r 100.00% 117.07%
750.sealcrypto_r 100.00% 92.00%
753.ns3_r 100.00% 91.87%
777.zstd_r 100.00% 97.16%
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 LTO LTO+PGO
706.stockfish_r 100.00% 96.72%
707.ntest_r 100.00% 94.76%
708.sqlite_r 100.00% 85.49%
710.omnetpp_r 100.00% 69.35%
714.cpython_r 100.00% 86.14%
721.gcc_r 100.00% 82.62%
723.llvm_r 100.00% 72.07%
727.cppcheck_r 100.00% 72.76%
729.abc_r 100.00% 71.02%
734.vpr_r 100.00% 75.73%
735.gem5_r 100.00% 71.13%
750.sealcrypto_r 100.00% 73.37%
753.ns3_r 100.00% 70.85%
777.zstd_r 100.00% 80.24%
Smaller binary sizes are always better.