Tensorflow Runs on GPU on home PC but not on Google colab

Question

I am trying to train the neural network in this repo on google colab: https://github.com/carpedm20/DCGAN-tensorflow

I'm running this code on Google Colab (with hardware accelerator set to GPU), and the training is 40 times slower than running on my home PC which has a much weaker GTX 1050. So I'm guessing on colab it is not running on the GPU. The output on google colab does indicate the GPU was found though:

Found GPU at: /device:GPU:0

Anyone know why this is?

Answer 1

i think the reason behind this data that that need time to come from driver to colab what about try GPU on parallel function 100 times in both that does not need data to run and tell me results

after all i found example that could finish first Epoch in 39 seconds when i try it it seems that colab take time to transfer data for each photo that i have and took in first Epoch 332s after initial things every-time you will run both he will get 39 seconds and i will get 332 s

i hope this help

Answer 2

In my case, there is a problem with loading data on Google Colab. It will strongly affect the speed even though we utilize the GPU.