Why does tensor flow use less than 20% of GEFORCE 1080 ti (11 GB)?

I am training the NMT model using the tf-seq2seq package from the GEFORCE GTX 1080 ti (11GB). During training, the model performing nvidia-smishows that the GPU's power consumption is always less than 25%, while all CPU cores are busy. Why?

nvidia-smi

How can I find if I can use the full potential of my GPU?

+4
source share
1 answer

The model in the above experiment was too small to use the full potential of the GPU.

( , , ...) ( 80-90%) .

+1

Source: https://habr.com/ru/post/1675076/


All Articles