October 2026 Summaries

No summary generated yet.

Vantage has launched NVIDIA Multi-Instance GPU cost accounting for Kubernetes workloads using NVIDIA’s single MIG strategy, allowing shared GPU costs to be allocated as fractions of the underlying physical GPU rather than treating each MIG slice as a full device. The capability automatically detects MIG slices and uses GPU Feature Discovery labels and cloud-provider GPU pricing to reconcile workload usage, costs, and idle capacity with the physical GPUs billed by AWS or Azure. For example, a workload using one of two equal MIG slices records 0.5 physical GPU-hours, while the unused half is assigned to the __idle__ namespace. The feature is included for all Vantage customers and requires Kubernetes Agent version 1.3.3 or later, enabled GPU metrics, and valid MIG labels; historical data can be backfilled when required labels are available. Mixed MIG configurations are not yet supported and retain existing full-GPU accounting, while GPU efficiency calculations remain unchanged.