Zesty launches automated fractional GPUs for Kubernetes

Zesty announced in a press release Automated Fractional GPUs, a capability that adjusts GPU memory allocation, compute sharing and workload placement in Kubernetes clusters. It follows the open source release of Run:ai by NVIDIA and works with its GPU Fractioning Operator.

The software uses GPU memory consumption, utilization and observed workload behavior to decide which workloads can share a GPU and how much memory each receives. It can change allocations as demand shifts while following performance, availability and resource policies set by teams.

Automated Fractional GPUs deploys as a lightweight agent and works with the Kubernetes scheduler. It supports time slicing and NVIDIA Multi Process Service, requires no application or model changes, and can be installed without restarting running workloads. The capability is available now through the Zesty platform.