Description
This container provides a robust foundation for AI workloads on NVIDIA GPUs with CUDA 12.9, delivering an enterprise-grade base image for Python libraries from the Red Hat Python Index. Built specifically for GPU-accelerated computing, it provides a secure and optimized environment for developing and deploying AI applications, with access to curated Python packages essential for machine learning, data science workflows, and modern AI applications.
- GPU-Accelerated Python Environment: The container provides a validated, accelerated stack to build containers for Red Hat AI, specifically optimized for NVIDIA GPU workloads with CUDA 12.9. It includes essential packages for machine learning and Python AI libraries to work with the Red Hat AI Python Index, all of which are validated and tested for compatibility with NVIDIA CUDA 12.9 environments.
- Optimized for NVIDIA Hardware: Specifically built and optimized for the NVIDIA CUDA 12.9 platform, this container enables you to leverage the full computational power of modern GPUs, including tensor cores and advanced GPU architectures. The environment is pre-configured to maximize performance for AI workloads running on NVIDIA hardware.
- Enterprise-Ready and Secure: Delivered as a hardened container based on Red Hat Enterprise Linux 9, it ensures the stability, security, and compliance required for mission-critical enterprise deployments. The container is an integral part of the Red Hat AI portfolio, offering a fully supported foundation for building and managing AI applications across all Red Hat AI products.
- Flexible Hybrid Cloud Deployment: Deploy consistently wherever you need it—in datacenters, clouds, or at the edge. This container is certified for and fully supported with Red Hat AI products. Use this container as the foundation for your AI infrastructure to build a powerful, standardized platform for GPU-accelerated workloads on NVIDIA hardware.
Leverage this container as the base layer for your AI applications, providing immediate access to enterprise-grade Python libraries while ensuring optimal performance on NVIDIA GPU infrastructure with CUDA 12.9.
If your goal is simply to serve a large language model, use instead the pre-built rhaiis/vllm-... containers, which are already optimized for this exact purpose.
General information
The following information was extracted from the containerfile and other sources.
| Summary | Red Hat AI Base Image for NVIDIA CUDA 12.9 on RHEL 9.6 |
| Description | Red Hat AI Base Image for NVIDIA CUDA 12.9 on RHEL 9.6 with index 3.3/cuda12.9-ubi9, Python 3.12, and AI repo 3.3 |
| Provider | Red Hat |
| Maintainer | Red Hat, Inc. |
Technical information
The following information was extracted from the containerfile and other sources.
| Repository name | rhai/base-image-cuda-12.9-rhel9 |
| Image version | 9.6 |
| Architecture | amd64 |
| Working directory | MPI_/usr/lib64/openmpi |
Use the following instructions to get images from a Red Hat container registry using registry service account tokens. You will need to create a registry service account to use prior to completing any of the following tasks.
Using OpenShift secrets
First, you will need to add a reference to the appropriate secret and repository to your Kubernetes pod configuration via an imagePullSecrets field.
Then, use the following from the command line or from the OpenShift Dashboard GUI interface.
Using podman login
Use the following command(s) from a system with podman installed
Using docker login
Use the following command(s) from a system with docker service installed and running
Use the following instructions to get images from a Red Hat container registry using your Red Hat login.
Using OpenShift
For best practices, it is recommended to use registry tokens when pulling content for OpenShift deployments.
Using podman login
Use the following command(s) from a system with podman installed
Using docker login
Use the following command(s) from a system with docker service installed and running
Getting source containers
Source code is available for all Red Hat UBI-based images in the form of downloadable containers. Here are a few things you should know about Red Hat source containers.
- Although they are packaged as containers, source containers cannot be run. So instead of using podman pull to get them to your system, use the skopeo command.
- Source containers are named based on the binary containers they represent. So, for example, to get the source container for a particular standard RHEL UBI 8 container (registry.access.redhat.com/ubi8/ubi8.1-397) you simply append -source to get the source code container for that image (registry.access.redhat.com/ubi8/ubi8.1-397-source).
- The skopeo command is recommended for getting source containers. With skopeo, you copy a source container to a directory on your local system for you to examine.
- Once a source container is copied to a local directory, you can use a combination of tar,gzip, and rpm commands to work with that content.
Step one
Use skopeo to copy the source image to a local directory
Step two
Inspect the image
Step three
Untar the contents
Step four
Begin examining and using the content.
