Back to Tachyon

MSM Benchmark

benchmark/msm/README.md

0.4.05.1 KB
Original Source

MSM Benchmark

CPU

Run on 13th Gen Intel(R) Core(TM) i9-13900K (32 X 5500 MHz CPU s)
Compiler: clang-15
CPU Caches:
  L1 Data 48 KiB (x16)
  L1 Instruction 32 KiB (x16)
  L2 Unified 2048 KiB (x16)
  L3 Unified 36864 KiB (x1)

Run on Apple M3 Pro (12 X 4050 MHz)
CPU Caches:
  L1 Data 64 KiB (x12)
  L1 Instruction 128 KiB (x12)
  L2 Unified 4096 KiB (x12)

Note: Run with build --@rules_rust//:extra_rustc_flags="-Ctarget-cpu=native" in your .bazelrc.user

Uniform points

shell
GOMP_SPINCOUNT=0 bazel run --config maxopt --//:has_matplotlib //benchmark/msm:msm_benchmark -- -k 16 -k 17 -k 18 -k 19 -k 20 -k 21 -k 22 -k 23 --vendor arkworks --vendor bellman --vendor halo2 --check_results

On Intel i9-13900K

ExponentTachyonArkworksBellmanHalo2
160.0284610.0377410.0774160.045742
170.0596480.0749360.1051040.08211
180.087430.127350.1966020.151715
190.1816460.2524240.3191850.282056
200.3038290.4545950.4710940.526231
210.5492870.9513970.8862441.00624
221.110212.007831.720111.9662
232.067623.784782.766733.68139

On Mac M3 Pro

ExponentTachyonArkworksBellmanHalo2
160.0460990.0517730.1108820.09505
170.0792980.0976980.1661830.174984
180.1519620.1736070.2968790.337657
190.2878480.341290.55630.592885
200.5049870.6304890.8409071.07097
210.9803021.333911.561961.98335
221.899772.867683.043923.9341
233.730155.714195.456367.51033

Non-uniform points

shell
GOMP_SPINCOUNT=0 bazel run --config maxopt --//:has_matplotlib //benchmark/msm:msm_benchmark -- -k 16 -k 17 -k 18 -k 19 -k 20 -k 21 -k 22 -k 23 --vendor arkworks --vendor bellman --vendor halo2 --test_set non_uniform --check_results

On Intel i9-13900K

ExponentTachyonArkworksBellmanHalo2
160.0301880.0336080.0577950.060642
170.0488510.0640590.1325840.099568
180.0801460.1215250.1241920.147496
190.1476260.2275170.2344290.27793
200.2896610.4451390.3411890.509375
210.4957070.8019750.7022591.0386
220.9937381.512661.248121.88462
231.699443.079042.000713.57452

On Mac M3 Pro

ExponentTachyonArkworksBellmanHalo2
160.0409540.0466630.0760680.089352
170.0699560.0893390.1193630.166812
180.1468690.1635780.2257680.326553
190.2684750.3024390.4600630.579915
200.5019560.6272720.7230711.09316
210.9207281.206621.223521.98457
221.789022.401242.245433.83765
233.479064.706514.133817.43978

GPU

Uniform points

shell
GOMP_SPINCOUNT=0 bazel run --config maxopt --//:has_matplotlib --config cuda //benchmark/msm:msm_benchmark_gpu -- -k 16 -k 17 -k 18 -k 19 -k 20 -k 21 -k 22 -k 23 --test_set non_uniform --check_results

On RTX-4090

ExponentTachyon CPUTachyon GPU
160.0266880.01981
170.0412910.006624
180.0814670.008306
190.1489290.012553
200.2608310.02423
210.4745420.044591
220.9212760.088349
231.702640.162646

Non-uniform points

shell
GOMP_SPINCOUNT=0 bazel run --config maxopt --//:has_matplotlib --config cuda //benchmark/msm:msm_benchmark_gpu -- -k 16 -k 17 -k 18 -k 19 -k 20 -k 21 -k 22 -k 23 --check_results

On RTX-4090

ExponentTachyon CPUTachyon GPU
160.0290450.020228
170.0475880.006565
180.0896730.008864
190.1648750.012308
200.291350.023396
210.5410670.043512
221.03790.08407
232.086010.157046