Red Hat and NVIDIA collaborate for optimised AI workloads

Red Hat, a leader in open source solutions, has unveiled support for its OpenShift platform on NVIDIA BlueField data processing units (DPUs). This new development aims to streamline the deployment of intensive AI application workloads with optimised security and enhanced networking and storage capabilities.

As modern business landscapes confront the dual challenge of processing vast data volumes with heightened security demands, AI applications often vie for resources with essential infrastructure services, potentially leading to performance issues and security vulnerabilities.

To counter these challenges, OpenShift on NVIDIA BlueField presents a unified, cloud-native platform. It introduces advanced networking, streamlined provisioning, and lifecycle management that effectively create a separation between AI applications and other infrastructure workloads, such as networking and security services.

• Optimised Resource Utilisation: OpenShift employs shrewd resource management, offloading networking services from the CPU to the DPU. This liberates resources for AI application workloads, maximising efficiency.
• Data Plane and Storage Traffic Acceleration: This integration speeds up data plane and storage traffic by shifting crypto and storage processing to the DPU, facilitating enhanced AI workload performance.
• Advanced Cloud Networking: With distributed routing functionalities powered by BlueField, OpenShift enhances secure multi-tenancy, network scalability, and performance while rationalising network operations.
• Improved Security Posture: By offloading tasks to BlueField, OpenShift reduces the attack surface, embracing a layered security approach to fortify application security.

Red Hat's support for the NVIDIA BlueField DPUs on OpenShift will soon be available as a technical preview, with plans to integrate more capabilities, including the NVIDIA DOCA software framework and third-party network functions.

NVIDIA's upcoming launch of the BlueField-4 is expected to bolster these capabilities further, heralding next-generation acceleration and performance. This development also promises to enhance capabilities within cloud-native AI environments, attributed to increased DOCA integration. Additionally, Red Hat’s collaboration with NVIDIA will facilitate support for Spectrum-X Ethernet networking, ensuring fluid, high-performance connectivity for AI workloads across cloud platforms.