Posted On: Mar 20, 2025
Today, NVIDIA's Multi-Instance GPU (MIG) feature support for Bottlerocket, the Linux-based operating system purpose-built for hosting containers, with a focus on security, minimal footprint, and safe updates, is now available in Amazon Web Services China Regions, enabling customers to partition NVIDIA GPUs into multiple GPU instances on Kubernetes nodes. This capability allows system administrators to maximize GPU resource utilization by running multiple workloads simultaneously on a single GPU while maintaining hardware-level isolation between workloads.
With MIG support, customers can optimize GPU resource allocation for workloads that don't fully utilize the GPU's compute capacity, such as machine learning inference tasks. Each GPU partition operates with complete hardware-level memory and fault isolation, providing workload separation and reliable performance.
NVIDIA Multi-Instance GPU support is available on all first-party Bottlerocket Amazon Machine Images starting with version 1.34.0 and is accessible in Amazon Web Services China (Beijing) Region, operated by Sinnet, and the Amazon Web Services China (Ningxia) Region, operated by NWCD. To learn more about MIG with the Bottlerocket NVIDIA variants, you can also visit the Bottlerocket GitHub repository for more information.
