SPEC CPU result comparison

Comparison of x86_64 levels - Summer 2022

SPECFP 2006: LTO -O2 - Different x86_64 levels

-flto -O2

AMD EPYC 7543P (zen3 based Milan)

Run-times

Benchmark x86-64 x86-64-v2 x86-64-v3 x86-64-v2 x86-64-v3
410.bwaves 100.00% 100.64% 105.08% 100.00% 104.42%
416.gamess 100.00% 101.07% 98.50% 100.00% 97.46%
433.milc 100.00% 97.90% 86.16% 100.00% 88.01%
434.zeusmp 100.00% 100.00% 96.36% 100.00% 96.36%
435.gromacs 100.00% 100.00% 90.93% 100.00% 90.93%
436.cactusADM 100.00% 99.59% 83.37% 100.00% 83.71%
437.leslie3d 100.00% 98.18% 96.97% 100.00% 98.77%
444.namd 100.00% 100.00% 89.43% 100.00% 89.43%
447.dealII 100.00% 100.61% 101.23% 100.00% 100.61%
450.soplex 100.00% 100.00% 98.50% 100.00% 98.50%
453.povray 100.00% 95.11% 93.89% 100.00% 98.72%
454.calculix 100.00% 99.53% 117.14% 100.00% 117.69%
459.GemsFDTD 100.00% 100.40% 92.12% 100.00% 91.75%
465.tonto 100.00% 99.40% 100.00% 100.00% 100.61%
470.lbm 100.00% 98.12% 89.75% 100.00% 91.47%
481.wrf 100.00% 100.00% 94.14% 100.00% 94.14%
482.sphinx3 100.00% 100.00% 97.35% 100.00% 97.35%
The compiler used was GCC 12.1.0.
Naturally, the smaller the run-times, the better. On the other hand, rates and especially their geometric means are better when they are greater.
See also continuous gcc tuning benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/tuning.

Non-debug executable sizes

Click to display/hide
Benchmark x86-64 x86-64-v2 x86-64-v3 x86-64-v2 x86-64-v3
410.bwaves 100.00% 100.00% 92.40% 100.00% 92.40%
416.gamess 100.00% 100.00% 99.96% 100.00% 99.96%
433.milc 100.00% 99.70% 98.44% 100.00% 98.73%
434.zeusmp 100.00% 100.00% 100.00% 100.00% 100.00%
435.gromacs 100.00% 99.69% 95.00% 100.00% 95.29%
436.cactusADM 100.00% 99.94% 97.56% 100.00% 97.62%
437.leslie3d 100.00% 99.98% 98.76% 100.00% 98.77%
444.namd 100.00% 99.89% 93.19% 100.00% 93.30%
447.dealII 100.00% 99.86% 99.96% 100.00% 100.10%
450.soplex 100.00% 100.13% 99.64% 100.00% 99.52%
453.povray 100.00% 99.37% 96.91% 100.00% 97.53%
454.calculix 100.00% 99.90% 95.43% 100.00% 95.53%
459.GemsFDTD 100.00% 99.76% 99.57% 100.00% 99.81%
465.tonto 100.00% 99.46% 97.99% 100.00% 98.52%
470.lbm 100.00% 99.76% 92.92% 100.00% 93.15%
481.wrf 100.00% 99.97% 99.18% 100.00% 99.21%
482.sphinx3 100.00% 100.12% 100.25% 100.00% 100.13%
Smaller binary sizes are always better.

Intel Cascade Lake Xeon

Run-times

Benchmark x86-64 x86-64-v2 x86-64-v3 x86-64-v2 x86-64-v3
410.bwaves 100.00% 100.00% 97.98% 100.00% 97.98%
416.gamess 100.00% 100.00% 94.70% 100.00% 94.70%
433.milc 100.00% 88.15% 87.75% 100.00% 99.54%
434.zeusmp 100.00% 99.60% 93.36% 100.00% 93.74%
435.gromacs 100.00% 98.97% 89.26% 100.00% 90.19%
436.cactusADM 100.00% 100.00% 87.65% 100.00% 87.65%
437.leslie3d 100.00% 100.61% 98.57% 100.00% 97.98%
444.namd 100.00% 100.40% 84.82% 100.00% 84.48%
447.dealII 100.00% 98.79% 97.59% 100.00% 98.78%
450.soplex 100.00% 100.00% 97.39% 100.00% 97.39%
453.povray 100.00% 100.00% 95.77% 100.00% 95.77%
454.calculix 100.00% 98.19% 99.20% 100.00% 101.02%
459.GemsFDTD 100.00% 100.00% 97.38% 100.00% 97.38%
465.tonto 100.00% 97.80% 96.99% 100.00% 99.18%
470.lbm 100.00% 100.00% 90.10% 100.00% 90.10%
481.wrf 100.00% 100.00% 94.15% 100.00% 94.15%
482.sphinx3 100.00% 99.79% 103.34% 100.00% 103.56%
The compiler used was GCC 12.1.0.
Naturally, the smaller the run-times, the better. On the other hand, rates and especially their geometric means are better when they are greater.
See also continuous gcc tuning benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/tuning.

Non-debug executable sizes

Click to display/hide
Benchmark x86-64 x86-64-v2 x86-64-v3 x86-64-v2 x86-64-v3
410.bwaves 100.00% 100.00% 92.40% 100.00% 92.40%
416.gamess 100.00% 100.00% 99.96% 100.00% 99.96%
433.milc 100.00% 99.70% 98.44% 100.00% 98.73%
434.zeusmp 100.00% 100.00% 100.00% 100.00% 100.00%
435.gromacs 100.00% 99.69% 95.00% 100.00% 95.29%
436.cactusADM 100.00% 99.94% 97.56% 100.00% 97.62%
437.leslie3d 100.00% 99.98% 98.76% 100.00% 98.77%
444.namd 100.00% 99.89% 93.19% 100.00% 93.30%
447.dealII 100.00% 99.86% 99.96% 100.00% 100.10%
450.soplex 100.00% 100.13% 99.64% 100.00% 99.52%
453.povray 100.00% 99.37% 96.91% 100.00% 97.53%
454.calculix 100.00% 99.90% 95.43% 100.00% 95.53%
459.GemsFDTD 100.00% 99.76% 99.57% 100.00% 99.81%
465.tonto 100.00% 99.46% 97.99% 100.00% 98.52%
470.lbm 100.00% 99.76% 92.92% 100.00% 93.15%
481.wrf 100.00% 99.97% 99.18% 100.00% 99.21%
482.sphinx3 100.00% 100.12% 100.25% 100.00% 100.13%
Smaller binary sizes are always better.