Closed denisb411 closed 6 years ago
IteratorGetNext
is part of your input pipeline, so it is running on CPU. You can get detail device placement information by setting --log_device_placement=True
.
So the training is indeed made on GPU (if avaible)?
@denisb411 you can check GPU utilization to confirm it is using GPU. You can also set --num_gpus=0
compare the speed with CPU only training.
Close for now due to no further activity.
Why is this happening? Training is too slow on CPU:0...
How can I set gpu:0 ?