How do you disable gpu from being utilized?

RLlib
1

How severe does this issue affect your experience of using Ray?

  • None

I have tried setting num_gpus=0, os.environ [‘CUDA_AVAILABLE_DEVICES’] = ‘-1’, os.environ[“RLLIB_NUM_GPUS”] = “0”, and even uninstalling cuda entirely. But for some reason, it still utilizes the gpu when training the agent. The environment is using PettingZoo with Pygame, render is disabled.

2

One thing you could try:

import os
os.environ[“CUDA_VISIBLE_DEVICES”] = “some_string_which_is_NOT_gpu_name”
from torch.cuda import device_count
print('Number of Devices: ', device_count())

I just use torch to count the devices, the CUDA_VISIBLE_DEVICES setting is not torch specific but should also work for tf.

3

I have tried using os.environ[“CUDA_VISIBLE_DEVICES”] but Ray still uses GPU vram. When I run the code on AMD GPU it doesn’t consume any vram.