Hey, so for some reason R crashes when I try to train an RNN (LSTM etc) on GPU, I know it works when training on CPU (from abck when I had no GPU). Is there a way to set the config such that keras is forced to train with CPU?
If you do the following you'll set CUDA cores to 0, hence disabling GPU indirectly:
# Disable GPU
Sys.setenv("CUDA_VISIBLE_DEVICES" = -1)
# Then load keras
library(keras)
use_backend("tensorflow")
Check this for further info: https://www.tensorflow.org/programmers_guide/using_gpu
I hope it helps
Most helpful comment
If you do the following you'll set CUDA cores to 0, hence disabling GPU indirectly:
Check this for further info: https://www.tensorflow.org/programmers_guide/using_gpu
I hope it helps