Benchmarks

Here we benchmark the model performance in two Architectures. The number of individuals used in the benchmark are (2^10, 2^15, 2^17, 2^20). And we also use different grid resolutions in 2-Dimensional and 3-Dimensional model setup.

0-Dimensional model

This is a benchmark of a simple 0-Dimensional model setup without advection of Eulerian tracers. However, the advection of individuals still take the same amount of time whether the velocity field is provided or not.

PlanktonIndividuals v0.6.1
Julia Version 1.8.0
Commit 5544a0fab76 (2022-08-17 13:38 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-13.0.1 (ORCJIT, broadwell)
  GPU: Tesla P100-PCIE-12GB
  CUDA runtime 11.8, artifact installation
  CUDA driver 11.2
  NVIDIA driver 460.84.0

Arch	N	min	median	mean	max	memory	allocs
CPU	1024	2.945 ms	3.016 ms	3.167 ms	4.328 ms	478.67 KiB	2992
CPU	32768	69.741 ms	69.812 ms	71.594 ms	80.231 ms	477.72 KiB	2931
CPU	131072	276.553 ms	276.966 ms	280.569 ms	300.907 ms	477.72 KiB	2931
CPU	1048576	2.582 s	2.590 s	2.590 s	2.598 s	477.72 KiB	2931
GPU	1024	7.085 ms	7.158 ms	7.364 ms	9.323 ms	1.92 MiB	21327
GPU	32768	7.435 ms	7.520 ms	7.925 ms	10.173 ms	1.92 MiB	21327
GPU	131072	7.053 ms	9.161 ms	9.851 ms	19.812 ms	1.92 MiB	21294
GPU	1048576	8.005 ms	46.217 ms	47.484 ms	122.516 ms	1.92 MiB	21294

2-Dimensional model

This is the benchmark of a 2-Dimensional model setup with (Ns, 1, Ns) grid cells. Here Ns = [32, 64, 128].

PlanktonIndividuals v0.6.1
Julia Version 1.8.0
Commit 5544a0fab76 (2022-08-17 13:38 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-13.0.1 (ORCJIT, broadwell)
  GPU: Tesla P100-PCIE-12GB
  CUDA runtime 11.8, artifact installation
  CUDA driver 11.2
  NVIDIA driver 460.84.0

Arch	N	Ns	min	median	mean	max	memory	allocs
CPU	1024	32	8.096 ms	8.132 ms	8.211 ms	8.688 ms	2.70 MiB	3109
CPU	1024	64	19.889 ms	19.940 ms	20.064 ms	20.952 ms	8.68 MiB	3052
CPU	1024	128	68.735 ms	69.030 ms	69.672 ms	75.046 ms	31.72 MiB	3052
CPU	32768	32	74.115 ms	74.154 ms	76.313 ms	85.288 ms	2.70 MiB	3048
CPU	32768	64	89.999 ms	90.163 ms	92.340 ms	101.475 ms	8.68 MiB	3052
CPU	32768	128	162.286 ms	162.618 ms	168.129 ms	190.011 ms	31.72 MiB	3052
CPU	131072	32	282.810 ms	282.913 ms	286.631 ms	307.620 ms	2.70 MiB	3048
CPU	131072	64	328.584 ms	328.962 ms	332.448 ms	357.787 ms	8.68 MiB	3052
CPU	131072	128	447.271 ms	453.263 ms	470.108 ms	509.040 ms	31.72 MiB	3052
CPU	1048576	32	2.476 s	2.476 s	2.501 s	2.552 s	2.70 MiB	3048
CPU	1048576	64	2.910 s	2.911 s	2.911 s	2.911 s	8.68 MiB	3052
CPU	1048576	128	2.905 s	2.909 s	2.909 s	2.914 s	31.72 MiB	3052
GPU	1024	32	6.902 ms	6.920 ms	7.101 ms	8.719 ms	1.98 MiB	21513
GPU	1024	64	7.417 ms	7.622 ms	7.755 ms	8.430 ms	2.07 MiB	21632
GPU	1024	128	7.734 ms	8.071 ms	8.141 ms	8.854 ms	2.45 MiB	21713
GPU	32768	32	7.011 ms	7.092 ms	7.392 ms	10.142 ms	1.98 MiB	21513
GPU	32768	64	6.769 ms	6.837 ms	7.152 ms	10.035 ms	2.07 MiB	21632
GPU	32768	128	7.027 ms	8.381 ms	8.561 ms	11.845 ms	2.45 MiB	21713
GPU	131072	32	6.580 ms	8.054 ms	8.560 ms	15.323 ms	1.98 MiB	21541
GPU	131072	64	7.491 ms	9.106 ms	9.664 ms	16.128 ms	2.07 MiB	21599
GPU	131072	128	7.918 ms	12.640 ms	12.791 ms	23.534 ms	2.45 MiB	21680
GPU	1048576	32	9.781 ms	35.539 ms	36.437 ms	59.171 ms	1.98 MiB	21528
GPU	1048576	64	10.682 ms	37.958 ms	39.055 ms	65.476 ms	2.08 MiB	21647
GPU	1048576	128	7.994 ms	50.094 ms	50.772 ms	126.537 ms	2.45 MiB	21680

3-Dimensional model

This is the benchmark of a 3-Dimensional model setup with (Ns, Ns, Ns) grid cells. Here Ns = [32, 64].

PlanktonIndividuals v0.6.1
Julia Version 1.8.0
Commit 5544a0fab76 (2022-08-17 13:38 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-13.0.1 (ORCJIT, broadwell)
  GPU: Tesla P100-PCIE-12GB
  CUDA runtime 11.8, artifact installation
  CUDA driver 11.2
  NVIDIA driver 460.84.0

Arch	N	Ns	min	median	mean	max	memory	allocs
CPU	1024	32	50.081 ms	50.249 ms	50.421 ms	51.994 ms	1.38 MiB	2820
CPU	1024	64	410.840 ms	459.105 ms	451.043 ms	459.516 ms	8.43 MiB	2821
CPU	32768	32	124.176 ms	124.312 ms	126.438 ms	138.224 ms	1.38 MiB	2820
CPU	32768	64	498.713 ms	534.237 ms	534.148 ms	554.501 ms	8.43 MiB	2821
CPU	131072	32	351.282 ms	351.674 ms	355.733 ms	387.071 ms	1.38 MiB	2820
CPU	131072	64	790.994 ms	808.337 ms	816.691 ms	848.149 ms	8.43 MiB	2821
CPU	1048576	32	3.019 s	3.072 s	3.072 s	3.125 s	1.38 MiB	2820
CPU	1048576	64	3.258 s	3.258 s	3.258 s	3.258 s	8.43 MiB	2821
GPU	1024	32	6.229 ms	6.286 ms	6.466 ms	7.329 ms	2.94 MiB	21053
GPU	1024	64	9.194 ms	11.891 ms	11.689 ms	12.604 ms	9.99 MiB	21077
GPU	32768	32	6.570 ms	6.638 ms	6.966 ms	8.974 ms	2.94 MiB	21053
GPU	32768	64	9.143 ms	12.882 ms	12.712 ms	15.781 ms	9.99 MiB	21077
GPU	131072	32	6.481 ms	9.150 ms	9.469 ms	16.907 ms	2.94 MiB	21081
GPU	131072	64	9.212 ms	16.623 ms	16.438 ms	25.557 ms	9.99 MiB	21105
GPU	1048576	32	7.257 ms	39.894 ms	40.268 ms	96.189 ms	2.94 MiB	21020
GPU	1048576	64	9.586 ms	54.934 ms	53.741 ms	118.675 ms	9.99 MiB	21105