Table of contents

Python deep learning libraries

Central processing unit (CPU)

The following Python deep learning libraries run on CPU only, and are preinstalled for you to use in notebooks:

Graphics processing unit (GPU)

DSX Local includes deep learning libraries from IBM PowerAI v1.5.1 which run on IBM POWER8 S822 LC and IBM POWER9 S822LC High Performance Computing systems with GPUs.

PowerAI includes the following GPU enabled libraries that can be used in a Jupyter notebook with a Python 2.7 kernel:

Additional libraries that are also supported:

To start a Jupyter notebook with PowerAI deep learning libraries, complete the following steps:

  1. Go to Environments.
  2. Find the environment Jupyter with Python 2.7 and PowerAI 1.5.1 for GPU for your project.
  3. Click Edit settings.
  4. Assign the desired number of GPUs, CPU and memory for your Jupyter container.
  5. Click Save and Start. Wait until the environment is started. Note that if there are not sufficient number of GPUs available, then the environment might remain in Pending state for a long time.
  6. When the environment has started, go to your project and open the desired notebook in the Jupyter with with Python 2.7 and PowerAI 1.5.1 for GPU environment.
  7. Run !nvidia-smi in the first cell to see the following output:

    | NVIDIA-SMI 396.26                 Driver Version: 396.26                    |
    | GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
    | Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Computer M.|
    |   0 Tesla P100-SMX2...  Off   | 00000002:01:00.0 Off |                    0 |
    | N/A   27C   P0   29W / 300W   |       0MiB / 1628MiB |      0%      Default |
    | Processes:                                                       GPU Memory |
    |  GPU       PID  Type  Process name                               Usage      |
    |  No running processes found                                                 |

You are now ready to run your PowerAI GPU accelerated program.

Restriction: Note that DSX Local for ICP only uses the GPUs available on the master node.