NVIDIA Donates DRA Driver for GPUs to Kubernetes Community to Advance Open Source AI

NVIDIA has donated its Dynamic Resource Allocation (DRA) driver for GPUs to the CNCF (Cloud Native Computing Foundation) to increase transparency and efficiency in high-performance AI infrastructure. Through this donation, the driver moves away from management by a single company and becomes fully community-owned under the Kubernetes project, which is expected to draw participation and innovation from a broader group of experts.

This driver offers developers the following key benefits:

• Improved efficiency: Supports NVIDIA Multi-Process Service (MPS) and NVIDIA Multi-Instance GPU (MIG) technologies to share GPU resources more intelligently.
• Large-scale scalability: Supports NVIDIA Multi-Node NVlink technology, enabling large-scale model training on next-generation AI infrastructure such as NVIDIA Grace Blackwell systems.
• Flexibility and precision: Hardware can be dynamically reconfigured, and applications can precisely request the specific compute power and memory configuration they need.

Additionally, NVIDIA has collaborated with the Kata Containers community to introduce GPU support, enabling AI workloads to run in security-enhanced Confidential Computing environments as well. This move aims to standardize the broader cloud native ecosystem through collaboration with industry leaders such as AWS, Google Cloud, Microsoft, and Red Hat.