CODEWITHSUNDEEP
pythontensorflowgpu
Marco
Marco

Reputation: 1235

Tensorflow GPU utilization below 10 %

I am trying to use my gpu NVIDIA GEFORCE GTX 1050, with tensorflow to train a neural network. I have tried with different code examples of different neural networks and the result is always a GPU utilization of 8% with computation that are much slower than with CPU. From this screen you can see the utilization during train.

enter image description here

Any solutions?

Upvotes: 1

Views: 2836

Answers (2)

Ivan Kovtun
Ivan Kovtun

Reputation: 693

Sometimes GPU utilization is bounded by HDD speed. Especially when training data does not fit into RAM and consists of millions of images I observed that one epoch training takes exactly the same time as reading all data from disk. Changing hdd with ssd or even nvm drive proportionally increased tensorflow training speed 5x in my case.

Upvotes: 3

kerastf
kerastf

Reputation: 509

It depends on your application.It is not unusual to have low GPU utilization.Try increasing the batch size

Upvotes: 1

Related Questions