Available types of Notebooks and consoles.
The following Notebooks are available.
Table 1. Notebook
types
| Notebook type |
Description |
| Python3 with PyTorch Elastic GPU Kernel |
The kernel starts with elastic distributed training for PyTorch. The GPU library is imported
but no GPU resource is requested. GPU allocation to the current kernel only happens when you
explicitly call the elastic distributed training train API in the Notebook and
request the maximum number. During training execution, GPUs are automatically allocated (between 1
GPU and the maximum number of GPUs). |
| Python3 with TensorFlow Elastic GPU Kernel |
The kernel starts with elastic distributed training for TensorFlow. The GPU library is
imported but no GPU resource is requested. GPU allocation to the current kernel only happens when
you explicitly call the elastic distributed training train API in Notebook and
request the maximum number. During training execution, GPUs are automatically allocated (between 1
GPU and the maximum number of GPUs). |
| Python3 with Single CPU Kernel |
The kernel starts with 1 CPU resource requested. More can be added. |
| Python3 with Single GPU Kernel |
The kernel starts with 1 GPU resource requested. More can be added. |
The following consoles are available.
Table 2. Console
types. The Jupyter console is a front-end terminal that uses Jupyter protocols for
running kernels.
| Console type |
Description |
| Python3 with PyTorch Elastic GPU |
A console that uses Python 3 with a PyTorch Elastic GPU kernel. |
| Python3 with TensorFlow Elastic GPU |
A console that uses Python 3 with a TensorFlow Elastic GPU Kernel. |
| Python3 with Single CPU |
A console that uses Python 3 with a Single CPU kernel. |
| Python3 with Single GPU |
A console that uses Python 3 with a Single GPU console kernel. |