GTX1080 LetNet-5 CPU GPU cuDNN5.1 时间对比

Caffe性能对比

最新推荐文章于 2022-01-04 17:29:00 发布

原创最新推荐文章于 2022-01-04 17:29:00 发布 · 1.2k 阅读

0 ·

CC 4.0 BY-SA版权

Caffe 专栏收录该内容

14 篇文章

订阅专栏

本文通过在CPU模式、GPU模式及cuDNN加速下运行LeNet模型进行对比测试，展示了不同配置下模型训练与测试的时间差异，揭示了GPU及cuDNN对于深度学习任务的重要作用。

CPU 模式：

Makefile.config

CPU_ONLY:=1

make clean
make -j4

./build/tools/caffe.bin time -model examples/mnist/lenet_train_test.prototxt
Testing for 50 iterations.

3350ms

GPU 模式：

Makefile.config

#CPU_ONLY:=1

make clean
make -j4

./build/tools/caffe.bin time -model examples/mnist/lenet_train_test.prototxt -gpu 0

Testing for 50 iterations.

I0105 10:14:01.645095 6806 caffe.cpp:412] Average Forward pass: 4.5064 ms.
I0105 10:14:01.645103 6806 caffe.cpp:414] Average Backward pass: 6.51756 ms.
I0105 10:14:01.645110 6806 caffe.cpp:416] Average Forward-Backward: 11.0997 ms.
I0105 10:14:01.645122 6806 caffe.cpp:418] Total Time: 554.985 ms.

cuDNN 模式：

Makefile.config

USE_CUDNN:=1

make clean
make -j

./build/tools/caffe.bin time -model examples/mnist/lenet_train_test.prototxt -gpu 0

Testing for 50 iterations.

I0105 10:20:37.588495 14720 caffe.cpp:412] Average Forward pass: 0.699844 ms.
I0105 10:20:37.588518 14720 caffe.cpp:414] Average Backward pass: 1.03538 ms.
I0105 10:20:37.588541 14720 caffe.cpp:416] Average Forward-Backward: 1.79493 ms.
I0105 10:20:37.588564 14720 caffe.cpp:418] Total Time: 89.7464 ms.