Article

NVIDIA has unveiled its Vera Rubin platform at GTC 2026, featuring seven new chips designed for next-gen AI datacenters. The Rubin GPU offers 288 GB HBM4 memory, 22 TB/s bandwidth, and 50 PFLOPs performance.

The NVL72 rack integrates 72 GPUs and 36 Vera CPUs, promising a 10x increase in inference throughput per watt compared to previous models. The platform also includes enhanced NVLink 6, ConnectX-9, and BlueField-4 technologies for improved efficiency. Vera Rubin aims to support advanced AI workloads, with shipments anticipated in late 2026.