According to the official description cited by TapTechNews, the GPU Partitioning technology aims to change the way virtual machines utilize GPU resources. Windows Server system can divide a single physical GPU into different partitions, and each partition can call a portion of the GPU performance.
Users can freely specify the GPU performance of Windows Server devices. Instead of requiring the entire GPU to be dedicated to a single virtual machine, it is now possible to share the GPU performance efficiently, as if each partition were an independent GPU.
This technology also uses a technology called failover clustering. If a virtual machine on a server node encounters a hardware failure or needs to be migrated, it can be restarted on another node in the cluster and use the GPU partition on another server.
Microsoft is also building a centralized management tool to make it easier for administrators to configure and monitor this new GPU virtualization setting. The Windows Admin Center user interface will provide a unified console for viewing detailed information about GPU partitions in the entire cluster environment and allocating these partitions to virtual machines as needed.