Tensorflow + Monero in Docker Impact?

[Biren78]

I know that Monero and Tensorflow use different CPU parts and Monero doesn't run full CPU.

Will mining Monero in Docker while training models in Tensorflow hurt training performance? A lot?

 

[Patrick]

Yes it will. Let me run something overnight so it is big enough to show a difference. Running a job on the machine in question now and will queue this after.

 

[Patrick]

Here is LeNet style MINST training eating CPU on a D-1540:

 

[Patrick]

LeNet style MINST training using Theano, with and without Monero mining in another container:

Train on 48000 samples, validate on 12000 samples
Epoch 1/20
48000/48000 [==============================] - 220s - loss: 0.1813 - acc: 0.9454 - val_loss: 0.0604 - val_acc: 0.9814
Epoch 2/20
48000/48000 [==============================] - 225s - loss: 0.0499 - acc: 0.9842 - val_loss: 0.0490 - val_acc: 0.9847
Epoch 3/20
48000/48000 [==============================] - 229s - loss: 0.0339 - acc: 0.9893 - val_loss: 0.0476 - val_acc: 0.9847
Epoch 4/20
48000/48000 [==============================] - 221s - loss: 0.0226 - acc: 0.9929 - val_loss: 0.0415 - val_acc: 0.9862
Epoch 5/20
48000/48000 [==============================] - 220s - loss: 0.0173 - acc: 0.9945 - val_loss: 0.0339 - val_acc: 0.9897
Epoch 6/20
48000/48000 [==============================] - 229s - loss: 0.0136 - acc: 0.9955 - val_loss: 0.0347 - val_acc: 0.9906
Epoch 7/20
48000/48000 [==============================] - 226s - loss: 0.0105 - acc: 0.9966 - val_loss: 0.0331 - val_acc: 0.9894
Epoch 8/20
48000/48000 [==============================] - 242s - loss: 0.0095 - acc: 0.9969 - val_loss: 0.0473 - val_acc: 0.9868
Epoch 9/20
48000/48000 [==============================] - 234s - loss: 0.0062 - acc: 0.9979 - val_loss: 0.0389 - val_acc: 0.9896
Epoch 10/20
48000/48000 [==============================] - 226s - loss: 0.0078 - acc: 0.9973 - val_loss: 0.0420 - val_acc: 0.9886
Epoch 11/20
48000/48000 [==============================] - 227s - loss: 0.0086 - acc: 0.9970 - val_loss: 0.0481 - val_acc: 0.9868
Epoch 12/20
48000/48000 [==============================] - 143s - loss: 0.0062 - acc: 0.9979 - val_loss: 0.0523 - val_acc: 0.9867
Epoch 13/20
48000/48000 [==============================] - 144s - loss: 0.0043 - acc: 0.9985 - val_loss: 0.0415 - val_acc: 0.9897
Epoch 14/20
48000/48000 [==============================] - 143s - loss: 0.0050 - acc: 0.9984 - val_loss: 0.0480 - val_acc: 0.9885
Epoch 15/20
48000/48000 [==============================] - 142s - loss: 0.0059 - acc: 0.9980 - val_loss: 0.0479 - val_acc: 0.9899

You can see the impact.