ML Runtimes NVIDIA RAPIDS Edition
The RAPIDS Edition Runtimes are built on top of community built RAPIDS docker images. The RAPIDS suite of software libraries gives you the freedom to execute end-to-end data science and analytics pipelines entirely on GPUs. It relies on NVIDIA CUDA primitives for low-level compute optimization, but exposes that GPU parallelism and high-bandwidth memory speed through user-friendly Python interfaces.
Visit rapids.ai for more information.
ML Runtimes RAPIDS edition differs from other Runtime Editions in the following ways:
- Python maintenance versions differ from what is being used in the Standard and NVIDIA GPU edition runtimes. These Python kernels are coming from the RAPIDS base image.
- Pre-installed Python packages and package versions differ from what’s in Standard and NVIDIA GPU edition runtimes.
The following RAPIDS editions are available for the 2021.04 Runtime version:
| RAPIDS and CUDA Version | Kernel | Editor | Base Image |
|---|---|---|---|
|
RAPIDS 0.18 CUDA 11.0 |
Python 3.7 | Workbench editor | rapidsai/rapidsai-core:0.18-cuda11.0-base-ubuntu20.04-py3.7 |
|
RAPIDS 0.18 CUDA 11.0 |
Python 3.8 | Workbench editor | rapidsai/rapidsai-core:0.18-cuda11.0-base-ubuntu20.04-py3.8 |
|
RAPIDS 0.18 CUDA 11.0 |
Python 3.7 | JupyterLab editor | rapidsai/rapidsai-core:0.18-cuda11.0-base-ubuntu20.04-py3.7Note: This image isn’t based on NVIDIA’s JupyterLab installation so RAPIDS library examples are not installed. |
|
RAPIDS 0.18 CUDA 11.0 |
Python 3.8 | JupyterLab editor |
|
RAPIDS Python environment
The RAPIDS python libraries are distributed and present as a conda environment in the docker image. While it’s possible to install additional libraries for your project via conda, these packages will be installed to a non-persistent part of the file system and won’t persist between sessions and jobs.
Installing additional libraries via pip is supported and works the same way as in other Runtimes.
