CalcPi Benchmark results
All tests were run on a PC with: Intel Core i5 2500K quad-core @ 4.2 GHz, 8GB RAM
Please be aware that this diagram is logarithmically scaled in both dimensions.
Click on this link to see the diagram full screen.
Contestants:
- y-cruncher 0.5.5
- y-cruncher 0.6.2 alpha
- CalcPi MSVC 64 1.0, original from the setup, using MPIR 2.6.0
- CalcPi MinGW64, with MPIR 2.6.0 optimized for AVX
- CalcPi GCC Linux 64, using GMP 5.0.2
CalcPi was set to the algorithm "PI: Parallel Chudnovsky (GMP)" with 4 threads.
Interpretation:
- For digits above 10'000'000, CalcPi is roughly 3 times slower than y-cruncher.
- The MPIR optimization for AVX (i.e. optimizing for --build=sandybridge-w64-mingw32 instead of --build=core2-w64-mingw32) only improves the speed by a few percent.
- GMP on Linux is equally fast as MPIR on Windows.
Download raw data of plot (as gnuplot script): gnuplot_bench.txt
Memory consumption:
- y-cruncher at 500'000'000 digits: 2.5GB
- CalcPi x64 MSVC Win8 at 250'000'000 digits: 2.2GB
- CalcPi x64 MSVC Win8 at 500'000'000 digits: 4.8GB