IBM Corp.’s Red Hat subsidiary today announced a series of updates intended to simplify and secure the deployment of artificial intelligence workloads across commercial, government and hybrid cloud environments.

These include support for Red Hat OpenShift on Nvidia BlueField data processing units, the introduction of a hardened container base image for use with Nvidia graphics processing units in government settings, and a new agreement to distribute the Nvidia Cuda Toolkit across the Red Hat software portfolio.

BlueField is a cloud-native platform that isolates infrastructure services, such as networking and security, from AI applications, reducing resource contention and improving overall performance and security. BlueField support for OpenShift offloads networking functions from the central processing unit to the DPU, accelerates storage traffic via support for Non-Volatile Memory Express over Fabrics and distributes routing to improve multitenant network scalability and security, Red Hat said.

The support will initially be available in a technical preview. Future releases are expected to deepen integration with Nvidia’s Data Center on a Chip Architecture software framework and support additional third-party network functions. Red Hat said it’s also working to integrate Nvidia Spectrum-X Ethernet technology to improve connectivity in distributed AI deployments.

CUDA across Red Hat platforms

As part of its ongoing collaboration with Nvidia, Red Hat will begin distributing the Nvidia CUDA Toolkit — a suite of tools for building GPU-accelerated applications — across its core software offerings, including Red Hat Enterprise Linux, OpenShift and Red Hat AI.

Red Hat said CUDA integration will reduce the operational complexity of AI deployments by allowing developers and information technology teams to obtain CUDA components directly from Red Hat’s repositories. The companies said they will also work toward supporting long-term interoperability between Red Hat platforms and future generations of Nvidia hardware and software.

Red Hat is also addressing the requirements of federal agencies and high-assurance environments by releasing a Security Technical Implementation Guide-hardened version of its Universal Base Image. This hardened container base image meets U.S. Department of Defense standards and is already being used by Nvidia to build a government-ready GPU Operator that runs on Red Hat OpenShift.

The hardened image is part of the Nvidia AI Factory for Government reference design, which is based on its Blackwell GPU platform and integrates Nvidia-certified systems and storage along with Red Hat technologies. The design supports agentic AI, physical AI and high-performance computing workloads across on-premise and hybrid cloud environments.

Red Hat said using UBI-STIG can reduce the time and manual effort required to achieve an Authority to Operate designation, improve consistency in software supply chains and run AI workloads on sensitive data more securely. The GPU Operator built on the image will be part of Nvidia AI Enterprise software, which is designed to meet Federal Risk and Authorization Management Program high-security requirements.

Photo: Paul Gillin/SiliconANGLE