Benchmark of CuArrays.jl

[GiggleLiu]

Platform:

CuArrays v0.8.0 [3a865a2d]
[0] NVIDIA: GeForce 940MX, 2004 MB
Intel(R) Core(TM) i5-7200U CPU @ 2.50GHz

Float32

julia> a = randn(Float32, 500, 500)
julia> af = cu(a)

MATMUL

julia> @benchmark $af*$af seconds=0.1
BenchmarkTools.Trial: 
  memory estimate:  400 bytes
  allocs estimate:  11
  --------------
  minimum time:     8.079 μs (0.00% GC)
  median time:      10.621 μs (0.00% GC)
  mean time:        125.724 μs (0.00% GC)
  maximum time:     476.358 μs (0.00% GC)
  --------------
  samples:          786
  evals/sample:     1

julia> @benchmark $a*$a seconds=0.1
BenchmarkTools.Trial: 
  memory estimate:  976.64 KiB
  allocs estimate:  2
  --------------
  minimum time:     2.404 ms (0.00% GC)
  median time:      3.594 ms (0.00% GC)
  mean time:        5.354 ms (0.00% GC)
  maximum time:     15.666 ms (0.00% GC)
  --------------
  samples:          20
  evals/sample:     1

FFT

julia> @benchmark fft($af)
BenchmarkTools.Trial: 
  memory estimate:  4.06 KiB
  allocs estimate:  101
  --------------
  minimum time:     927.302 μs (0.00% GC)
  median time:      974.301 μs (0.00% GC)
  mean time:        1.972 ms (35.31% GC)
  maximum time:     118.619 ms (71.07% GC)
  --------------
  samples:          2553
  evals/sample:     1

julia> @benchmark fft($a)
BenchmarkTools.Trial: 
  memory estimate:  3.82 MiB
  allocs estimate:  57
  --------------
  minimum time:     4.199 ms (0.00% GC)
  median time:      4.452 ms (0.00% GC)
  mean time:        4.729 ms (4.61% GC)
  maximum time:     67.616 ms (91.65% GC)
  --------------
  samples:          1056
  evals/sample:     1

ComplexF64

julia> a = randn(ComplexF64, 500, 500)
julia> af = cu(a)

MATMUL

julia> @benchmark $af*$af seconds=0.1
BenchmarkTools.Trial: 
  memory estimate:  400 bytes
  allocs estimate:  11
  --------------
  minimum time:     22.125 μs (0.00% GC)
  median time:      23.201 μs (0.00% GC)
  mean time:        180.813 μs (82.23% GC)
  maximum time:     83.866 ms (97.15% GC)
  --------------
  samples:          548
  evals/sample:     1

julia> @benchmark $a*$a seconds=0.1
BenchmarkTools.Trial: 
  memory estimate:  3.81 MiB
  allocs estimate:  2
  --------------
  minimum time:     16.390 ms (0.00% GC)
  median time:      16.736 ms (0.00% GC)
  mean time:        20.200 ms (15.91% GC)
  maximum time:     35.663 ms (54.06% GC)
  --------------
  samples:          6
  evals/sample:     1

FFT

julia> @benchmark fft($af) seconds=0.1
BenchmarkTools.Trial: 
  memory estimate:  704 bytes
  allocs estimate:  15
  --------------
  minimum time:     762.266 μs (0.00% GC)
  median time:      788.589 μs (0.00% GC)
  mean time:        831.032 μs (0.00% GC)
  maximum time:     1.237 ms (0.00% GC)
  --------------
  samples:          120
  evals/sample:     1

julia> @benchmark fft($a) seconds=0.1
BenchmarkTools.Trial: 
  memory estimate:  3.82 MiB
  allocs estimate:  55
  --------------
  minimum time:     4.197 ms (0.00% GC)
  median time:      5.443 ms (0.00% GC)
  mean time:        11.459 ms (55.55% GC)
  maximum time:     63.728 ms (91.32% GC)
  --------------
  samples:          12
  evals/sample:     1