CoreWeave brings up multi-rack NVIDIA Vera Rubin NVL72 clusters, shares engineering details

OnlineInference · x · 2026-09-17

CoreWeave announced the bring-up of multi-rack NVIDIA Vera Rubin NVL72 clusters — each rack packs 72 Rubin GPUs, 36 Vera CPUs, NVLink 6 fabric, and 45°C liquid cooling, with hundreds of GPUs networked into one scale-out system for training, inference, and agentic AI workloads. The blog details multi-rack engineering pitfalls: straggler GPUs, marginal network links hurting collectives, cooling anomalies, and config consistency across NVLink domains.

Related event: CoreWeave Lights Up Multi-Rack NVIDIA Vera Rubin NVL72 Clusters(2 posts)→

Original post →

More from Infra

Infra channel →