Skip to content

Instantly share code, notes, and snippets.

@mwrnd

mwrnd/benchmark_results_RX580_ROCm1.9.3

Created October 23, 2019 20:19
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save mwrnd/e28ab9b935fe79f78df7806854d66772 to your computer and use it in GitHub Desktop.
Save mwrnd/e28ab9b935fe79f78df7806854d66772 to your computer and use it in GitHub Desktop.
Download ZIP
AMD Radeon RX 580 Tensorflow benchmarking results with ROCm 1.9.3
Raw
benchmark_results_RX580_ROCm1.9.3
AMD Radeon RX 580 8GB tensorflow/benchmarks Results
By Matthew Wielgus 2019-10-22
Video Card: MSI Radeon RX 580 8GB ARMOR OC (rocm-smi -v Cannot get VBIOS version)
Motherboard: MSI X570-A Pro with 32GB DDR4-2133 BIOS H.40
Processor: AMD Ryzen 5 3600X
OS: Ubuntu 18.04.0 no apt upgrade or apt dist-upgrade
Kernel: 4.15.0-20-generic
rocm-dkms: 1.9.3 installed through apt
tensorflow-rocm: 1.12.0 installed through pip
tensorflow benchmarks: 091ef1e4d8832e14d1f874e66bff78a2522d0947
https://github.com/tensorflow/benchmarks/archive/091ef1e4d8832e14d1f874e66bff78a2522d0947.zip
tensorflow_models: 1.12.0
https://codeload.github.com/tensorflow/models/zip/v1.12.0
Legend:
X means XLA was enabled
export TF_XLA_FLAGS=--tf_xla_cpu_global_jit
R means ROCm Fusion was enabled
export TF_ROCM_FUSION_ENABLE=1
F means 16-Bit Floating Point was used (--use_fp16)
C means MIOpen Compute Unit optimations do not exist for this GPU
na means the benchmark did not run or batch size was too large
GPU
imagenet
model batchsize= 016 016F 016R 016RF 032 032F 032R 032RF 064 064F 064R 064RF 128 128F 128R 128RF 256 256F 256R 256RF
alexnet 256 188 259 188 401 262 402 263 511 327 529 327 627 376 650 374 710 396 725 396
googlenet 209 119 219 120 247 136 246 142 261 147 261 153 271 153 na 157 248 na 257 159
inception3 47.0 22.9 48.6 23.2 48.9 23.8 50.4 24.2 51.6 na 53.3 24.6 na na na 24.5 na na na na
inception4 21.1 10.8 na 11.1 23.0 11.2 24.0 11.4 11.4 na na na na na na na na na na na
lenet5 3670 3118 na 3114 6238 5354 6336 5316 10041 8310 10273 8319 15455 11948 15520 11956 20690 15770 20767 15756
mobilenet 328 298 529 314 434 264 954 482 304 na 1263 670 485 370 1590 819 513 408 1836 920
nasnet 8.0 8.5 5.3 6.9 na 8.8 4.7 7.5 8.8 8.4 na 7.0 na na na na na na na na
overfeat 92.7 55.8 na 55.8 145 75.1 145 75.7 201 91.2 201 91.0 229 101 247 101 275 na 274 na
resnet101 na 46.7 na 50.1 54.5 30.3 58.5 31.9 32.0 na na na na na na na na na na na
resnet101_v2 47.6 27.8 na 28.4 55.2 31.4 59.9 32.3 33.0 na na na na na na na na na na na
resnet152 33.5 18.9 na 19.5 38.2 21.2 41.2 21.7 22.3 na na na na na na na na na na na
resnet152_v2 33.8 19.0 na 19.6 38.5 21.3 41.9 21.9 22.4 na na na na na na na na na na na
resnet50 78.6 51.7 na 53.3 92.0 91.9 99.1 59.4 100 61.1 112 63.0 60.7 na na na na na na na
resnet50_v1.5 62.3 46.4 na 48.6 75.7 51.5 81.7 53.7 na 54.3 86.1 56.6 56.6 na na na na na na na
resnet50_v2 79.6 52.1 na 54.0 93.6 58.2 102 60.3 106 61.7 114 63.9 63.6 na na na na na na na
trivial 3869 527 na 527 7012 1038 6861 1038 11879 2039 12030 2038 18737 3918 18628 3919 26770 7281 26870 7292
vgg11 64.7 35.1 68.0 35.4 82.7 39.3 81.4 39.2 92.9 41.5 94.5 41.5 95.5 41.9 101 41.8 41.7 na na na
vgg16 35.5 18.8 37.2 18.8 39.9 20.2 42.2 20.2 44.6 20.8 45.7 20.8 21.0 na na na na na na na
vgg19 31.2 15.1 na 15.0 35.6 16.1 35.6 16.1 37.8 16.5 37.7 16.5 16.7 na na na na na na na
Average Gain -37.9% +6.7% -37.8% -40.1% +7.7% -39.3% -42.7% +27.9% -32.2% -45.9% +40.8% -35.0% -40.3% +44.0% -19.5%
Median Gain -42.3% +4.8% -41.1% -44.5% +4.4% -42.5% -42.7% +2.5% -40.6% -43.5% +4.7% -47.3% -34.0% +1.2% -35.9%
cifar10
model batchsize= 016 016F 016R 016RF 032 032F 032R 032RF 064 064F 064R 064RF 128 128F 128R 128RF 256 256F 256R 256RF
alexnet 3021 232 na 234 4782 248 4931 250 6704 257 6707 257 8322 264 8317 264 9838 268 9911 267
nasnet na 42.0 na 8.7 47.9 49.8 5.1 19.7 53.7 56.4 na 10.6 55.3 57.7 na na na na na na
resnet110 462 260 na 273 671 407 731 434 852 549 944 580 852 618 922 639 912 669 984 687
resnet110_v2 465 261 na na 674 409 na na 848 550 na na 847 617 na na 913 668 na na
resnet20 2075 1266 na 2222 2988 1993 3147 2093 3799 2745 4073 2865 3947 3121 4179 3226 4313 3406 4560 3470
resnet20_v2 2035 1254 na na 2902 1990 na na 3674 2723 na na 3810 3098 na na 4197 3358 na na
resnet32 1437 843 na 1570 2062 1320 2245 1385 2616 1800 2837 1881 2664 2035 2844 2096 2888 2202 3074 2259
resnet32_v2 1442 846 na na 2048 1310 na na 2566 1795 na na 2615 2024 na na 2839 2187 na na
resnet44 1090 633 na 1180 1571 979 1714 1031 1988 1336 2167 1403 2005 1508 2157 1552 2170 1631 2321 1671
resnet44_v2 1094 627 na na 1558 978 na na 1960 1334 na na 1981 1498 na na 2141 1619 na na
resnet56 872 499 na 962 1272 785 1367 826 1604 1062 1745 1118 1612 1195 1736 1234 1739 1294 1862 1326
resnet56_v2 881 501 na na 1253 789 na na 1585 1062 na na 1598 1192 na na 1719 1288 na na
trivial 7965 4236 na 7216 15422 8041 15390 8392 25992 15631 22976 15648 46789 32040 46288 29817 71242 56869 54473 54308
Average Gain -46.4% -15.4% -38.8% -5.8% -45.8% -34.5% +4.7% -45.1% -28.1% +5.0% -34.8% -29.9% +1.6% -33.4%
Median Gain -42.7% +7.1% -37.2% +6.4% -35.2% -32.8% +8.5% -31.1% -24.8% +6.8% -23.5% -24.6% +6.4% -23.8%
CPU
cifar10
model batchsize= 016 016F 016R 016RF 032 032F 032R 032RF 064 064F 064R 064RF 128 128F 128R 128RF 256 256F 256R 256RF
trivial 14859 945 na na 22224 1050 na na 22525 1914 na na 37755 2504 na na 42527 2581 na na
imagenet
model batchsize= 016 016F 016R 016RF 032 032F 032R 032RF 064 064F 064R 064RF 128 128F 128R 128RF 256 256F 256R 256RF
mobilenet 113 5.9 na na 132 7.0 na na 140 7.1 na na 144 na na na na 146 na na
trivial 938 50.6 na na 1227 52.6 na na 1387 52.6 na na 1789 53.0 na na 1898 53.1 na na
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment