Running GPU from Docker containers

Training predictive models, especially neural networks with many parameters takes lots of computing power. To use a GPU the container must be configured to access the GPU from the container and contain drivers and libraries. The container in https://github.com/learniotai/iotai_container_env is configured to use Nvidia GPUs. Currently Tensorflow and Pytorch are built to use only Nvidia GPUs through the CUDA application programming interface (API) though support for AMD Radeon GPUs is in progress for Pytorch.

Setup Docker and computing environment.

Launch the Docker container with GPU support.

iotai_train

If you have a GPU check that it’s visible from inside the Docker image.

$ learniotai@546e6389eba7:/opt$ nvidia-smi 
Wed Apr 28 02:10:39 2021       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 460.56       Driver Version: 460.56       CUDA Version: 11.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  GeForce GTX 1070    Off  | 00000000:01:00.0  On |                  N/A |
| 27%   36C    P8    12W / 151W |    227MiB /  8119MiB |     38%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
+-----------------------------------------------------------------------------+

Test machine learning tool kits can access GPU

Test Tensorflow GPU

$ python
import tensorflow as tf
tf.test.is_gpu_available()
...
StreamExecutor device (0): GeForce GTX 1070, Compute Capability 6.1
True

Test Pytorch GPU

import torch 
torch.cuda.is_available()
True