Keras：CuDNN 不可用？

Question

我是运行 Keras 的 example kaggle_otto_nn.py，后端是 theano。

在下面的打印输出中，第 5 行说：

CNMeM is enabled with initial size: 90.0% of memory, CuDNN not available

我想知道，这 CuDNN not available 重要吗，因为 GPU 设备是可检测的？我运行我在 GPU 上的程序正确吗？或者它实际上不是运行在 GPU 上。

cliu@cliu-ubuntu:keras-examples$ THEANO_FLAGS=mode=FAST_RUN,device=gpu,floatX=float32,lib.cnmem=0.9 python kaggle_otto_nn.py
Using Theano backend.
/usr/local/lib/python2.7/dist-packages/Theano-0.8.0rc1-py2.7.egg/theano/tensor/signal/downsample.py:6: UserWarning: downsample module has been moved to the theano.tensor.signal.pool module.
  "downsample module has been moved to the theano.tensor.signal.pool module.")
Using gpu device 0: Quadro K610M (CNMeM is enabled with initial size: 90.0% of memory, CuDNN not available)
Loading data...
9 classes
93 dims
Building model...
Training model...
Train on 52596 samples, validate on 9282 samples
Epoch 1/20
52596/52596 [==============================] - 6s - loss: 0.9420 - val_loss: 0.6269
Epoch 2/20
52596/52596 [==============================] - 6s - loss: 0.6955 - val_loss: 0.5817
...
Epoch 20/20
52596/52596 [==============================] - 6s - loss: 0.4866 - val_loss: 0.4819
Generating submission...
144368/144368 [==============================] - 1s     
Wrote submission to file keras-otto.csv.

Answer 1

cuDNN 是 NVidia 的一个库，它提高了神经网络在 GPU 上的性能。所以你的程序仍然在 GPU 上运行，但比安装 cuDNN 时慢得多。

Keras：CuDNN 不可用？

Keras: CuDNN not available?

python

gpu

python-2.7

theano

keras