CoreWeave is now offering Nvidia GB200 NVL72 instances on its cloud platform.
The instances are generally available via CoreWeave's Kubernetes Service, Slurm oN Kubernetes (SUNK), and Mission Control platform.
The GB200 NVL72-based instances on CoreWeave connect 36 Nvidia Grace CPUs and 72 Nvidia Blackwell GPUs in a liquid-cooled, rack-scale design and are available as bare-metal instances through CoreWeave Kubernetes Service, and are scalable up to 110,000 GPUs.
The instances use Nvidia Quantum-2 InfiniBand networking with 400Gbps of bandwidth per GPU, and deploy the chip vendor's Quantum-2's SHARP In-Network Computing technology to reduce latency further.
The instances are currently available via the US-West-01 region.
CoreWeave claims that it is the first cloud provider to make Nvidia GB200 NVL72 instances generally available.
The company displayed a demonstration of a GB200 NVL72 system at one of its data centers in November of last year. The cluster delivered up to 1.4 exaFLOPS of AI compute.
In January 2025, it was reported that IBM would be using CoreWeave's cloud platform to access the Nvidia GB200 clusters, featuring the GB200 NVL72 systems for training its next generation of Granite AI models.
Last month, Lambda deployed two GB200 NVL72 racks - one at an EdgeCloudLink data center and another at a Pegatron facility. Microsoft was the first cloud provider to deploy Nvidia’s new GB200 GPUs in its AI cloud servers, although, according to reports, that rack configuration was not a GB200 NVL72. Google also gave a sneak peek of its upcoming Nvidia Blackwell GB200 NVL racks in October 2024.
