矩池云 RTX 2080 Ti+Ubuntu18.04+Tensorflow1.15.2 性能测试！
今天为了对比滴滴云(大师码：8888) NVIDIA A100，特地跑了一下RTX2080的TensorFlow基准测试，现在把结果记录一下！
平台为：矩池云
系统为：Ubuntu 18.04
显卡为：RTX 2080 Ti
Python版本： 3.6.10
TensorFlow版本：1.15.2
显卡相关内容如下：
系统配置如下：
测试方法：
https://github.com/tensorflow/benchmarks
Resnet50 BS64
python tf_cnn_benchmarks.py --num_gpus=1 --batch_size=64 --model=resnet50
Step Img/sec total_loss 1 images/sec: 305.5 +/- 0.0 (jitter = 0.0) 8.220 10 images/sec: 305.2 +/- 0.3 (jitter = 0.7) 7.880 20 images/sec: 305.3 +/- 0.2 (jitter = 0.9) 7.910 30 images/sec: 305.1 +/- 0.2 (jitter = 0.8) 7.820 40 images/sec: 304.9 +/- 0.2 (jitter = 0.7) 8.005 50 images/sec: 304.8 +/- 0.1 (jitter = 0.9) 7.770 60 images/sec: 304.5 +/- 0.2 (jitter = 1.1) 8.114 70 images/sec: 304.3 +/- 0.2 (jitter = 1.3) 7.816 80 images/sec: 304.2 +/- 0.2 (jitter = 1.5) 7.975 90 images/sec: 304.0 +/- 0.1 (jitter = 1.5) 8.094 100 images/sec: 303.8 +/- 0.1 (jitter = 1.6) 8.035 ---------------------------------------------------------------- total images/sec: 303.65 ----------------------------------------------------------------
AlexNet BS512
python tf_cnn_benchmarks.py --num_gpus=1 --batch_size=512 --model=alexnet
Step Img/sec total_loss 1 images/sec: 3939.5 +/- 0.0 (jitter = 0.0) nan 10 images/sec: 3927.5 +/- 3.0 (jitter = 12.2) nan 20 images/sec: 3923.9 +/- 2.1 (jitter = 11.7) nan 30 images/sec: 3923.0 +/- 2.5 (jitter = 11.0) nan 40 images/sec: 3921.2 +/- 2.0 (jitter = 9.4) nan 50 images/sec: 3919.0 +/- 1.8 (jitter = 9.2) nan 60 images/sec: 3915.4 +/- 1.9 (jitter = 11.5) nan 70 images/sec: 3912.2 +/- 2.0 (jitter = 13.7) nan 80 images/sec: 3911.5 +/- 1.8 (jitter = 14.5) nan 90 images/sec: 3909.8 +/- 1.8 (jitter = 15.9) nan 100 images/sec: 3907.9 +/- 1.7 (jitter = 15.9) nan ---------------------------------------------------------------- total images/sec: 3905.13 ----------------------------------------------------------------
Inception v3 BS64
python tf_cnn_benchmarks.py --num_gpus=1 --batch_size=64 --model=inception3
Step Img/sec total_loss 1 images/sec: 200.6 +/- 0.0 (jitter = 0.0) 7.278 10 images/sec: 200.6 +/- 0.1 (jitter = 0.6) 7.298 20 images/sec: 200.5 +/- 0.1 (jitter = 0.4) 7.291 30 images/sec: 200.3 +/- 0.1 (jitter = 0.4) 7.412 40 images/sec: 200.1 +/- 0.1 (jitter = 0.7) 7.306 50 images/sec: 199.9 +/- 0.1 (jitter = 0.8) 7.287 60 images/sec: 199.7 +/- 0.1 (jitter = 1.0) 7.378 70 images/sec: 199.5 +/- 0.1 (jitter = 1.2) 7.351 80 images/sec: 199.3 +/- 0.1 (jitter = 1.3) 7.402 90 images/sec: 199.2 +/- 0.1 (jitter = 1.2) 7.309 100 images/sec: 199.0 +/- 0.1 (jitter = 1.2) 7.354 ---------------------------------------------------------------- total images/sec: 198.97 ----------------------------------------------------------------
VGG16 BS64
python tf_cnn_benchmarks.py --num_gpus=1 --batch_size=64 --model=vgg16
Step Img/sec total_loss 1 images/sec: 180.0 +/- 0.0 (jitter = 0.0) 7.346 10 images/sec: 179.5 +/- 0.1 (jitter = 0.2) 7.294 20 images/sec: 179.4 +/- 0.1 (jitter = 0.3) 7.282 30 images/sec: 179.1 +/- 0.1 (jitter = 0.4) 7.278 40 images/sec: 178.9 +/- 0.1 (jitter = 0.8) 7.287 50 images/sec: 178.7 +/- 0.1 (jitter = 0.7) 7.272 60 images/sec: 178.6 +/- 0.1 (jitter = 0.7) 7.261 70 images/sec: 178.4 +/- 0.1 (jitter = 1.0) 7.267 80 images/sec: 178.3 +/- 0.1 (jitter = 1.1) 7.280 90 images/sec: 178.2 +/- 0.1 (jitter = 1.0) 7.270 100 images/sec: 178.1 +/- 0.1 (jitter = 0.9) 7.268 ---------------------------------------------------------------- total images/sec: 178.02 ----------------------------------------------------------------
GoogLeNet BS128
python tf_cnn_benchmarks.py --num_gpus=1 --batch_size=128 --model=googlenet
Step Img/sec total_loss 1 images/sec: 784.7 +/- 0.0 (jitter = 0.0) 7.104 10 images/sec: 782.9 +/- 0.4 (jitter = 1.4) 7.104 20 images/sec: 782.3 +/- 0.6 (jitter = 2.1) 7.092 30 images/sec: 780.3 +/- 0.7 (jitter = 4.3) 7.087 40 images/sec: 779.2 +/- 0.6 (jitter = 5.5) 7.067 50 images/sec: 778.9 +/- 0.5 (jitter = 5.0) 7.092 60 images/sec: 778.4 +/- 0.5 (jitter = 4.7) 7.050 70 images/sec: 778.3 +/- 0.4 (jitter = 4.2) 7.073 80 images/sec: 778.2 +/- 0.4 (jitter = 3.9) 7.077 90 images/sec: 778.2 +/- 0.4 (jitter = 3.0) 7.079 100 images/sec: 778.1 +/- 0.3 (jitter = 2.7) 7.066 ---------------------------------------------------------------- total images/sec: 777.65 ----------------------------------------------------------------
ResNet152 BS32
python tf_cnn_benchmarks.py --num_gpus=1 --batch_size=32 --model=resnet152
Step Img/sec total_loss 1 images/sec: 116.5 +/- 0.0 (jitter = 0.0) 9.028 10 images/sec: 116.3 +/- 0.1 (jitter = 0.2) 8.593 20 images/sec: 116.2 +/- 0.1 (jitter = 0.3) 8.603 30 images/sec: 116.0 +/- 0.1 (jitter = 0.4) 8.712 40 images/sec: 115.8 +/- 0.1 (jitter = 0.5) 8.655 50 images/sec: 115.7 +/- 0.1 (jitter = 0.6) 8.800 60 images/sec: 115.7 +/- 0.1 (jitter = 0.6) 8.625 70 images/sec: 115.5 +/- 0.1 (jitter = 0.6) 9.093 80 images/sec: 115.5 +/- 0.1 (jitter = 0.6) 8.856 90 images/sec: 115.4 +/- 0.1 (jitter = 0.6) 8.996 100 images/sec: 115.3 +/- 0.1 (jitter = 0.6) 8.842 ---------------------------------------------------------------- total images/sec: 115.28 ----------------------------------------------------------------
A100 和V100 和 2080ti 性能对比：
https://www.tonyisstark.com/383.html
如果觉得我的文章对您有用，请随意打赏。您的支持将鼓励我继续创作！
