Kubernetes optimization leader Zesty has launched Automated Fractional GPUs to bring autonomous resource allocation to NVIDIA CUDA-native GPU fractioning, enabling enterprises to maximize utilization and curb AI infrastructure expenses. Built on NVIDIA’s open-source release of Run:ai on the NVIDIA DSX AI Factory Platform, Zesty adds a continuous, intelligent decision layer that dynamically adjusts GPU memory allocation, compute sharing modes, and workload placement based on real-time application demand, latency requirements, and set performance limits. Because traditional static partitioning methods like time-slicing, MPS, or MIG often leave expensive GPU capacity underutilized, Zesty’s solution deploys as a lightweight Kubernetes agent alongside the NVIDIA Run:ai GPU Fractioning Operator, requiring zero code modifications or application restarts.
Also Read: MaxLinear Debuts Carmel High-Speed USB-to-UART Solution to Accelerate AI Data Center Infrastructure Management
Highlighting the strategic value of this release, Pini Ben Nahum, VP R&D at Zesty, stated, “GPUs are among the most expensive and constrained resources infrastructure teams manage today. They’re difficult to secure, costly to add, and shared by AI workloads whose needs change constantly. Now NVIDIA has opened GPU fractioning to the ecosystem, and Zesty is making their allocation intelligent and autonomous,” By intelligently driving workload density and rightsizing resource usage without adding operational risk or manual labor, Automated Fractional GPUs is now available as part of the core Zesty platform to help AI teams extract more performance from their existing hardware investments.
