Hyperscaler CoreWeave has made a notable advancement by becoming the first AI cloud provider to deploy Nvidia’s latest GB300 NVL72 systems. This milestone was achieved through collaboration with Dell Technologies, Switch, and Vertiv, aimed at addressing the rigorous demands of AI reasoning.
The new CoreWeave GB300 NVL72 setup is a rack-scale, liquid-cooled platform that integrates 72 Nvidia Blackwell Ultra GPUs, 36 Arm-based Nvidia Grace CPUs, and 36 NVIDIA BlueField-3 DPUs. This configuration is designed to enhance CoreWeave’s cloud-native software offerings, including the CoreWeave Kubernetes Service (CKS) and Slurm on Kubernetes (SUNK).
Dell highlighted that CoreWeave’s use of Integrated Racks equipped with Nvidia GB300 NVL72 sets a new standard for scalability in cloud services, facilitating tasks such as extensive language model training, reasoning, and real-time inference.
CoreWeave’s Chief Technology Officer, Peter Salanki, noted that this system is specifically engineered to cater to the high computational needs of test-time scaling inference, essential for the deployment of cutting-edge AI models. He emphasized that as AI models continue to evolve in complexity and size, the demand for specialized AI infrastructure will increase significantly.
This deployment marks a significant achievement for Dell, according to Matt Kimball, Vice President and Principal Analyst at Moor Insights & Strategy. He described it as beneficial for setting market pace and enhancing the relationship between Dell and Nvidia. Kimball remarked on the evolution of Dell’s strategy in the cloud market after previously stepping back in 2016, noting their recent focus on integrated systems and commitments to quality.
Despite earlier concerns regarding overheating issues with Nvidia’s Blackwell processors, Kimball expressed confidence in Nvidia’s careful product development and Dell’s commitment to quality control.
CoreWeave’s primary focus remains on delivering GPU capabilities to a market heavily interested in AI. Clients such as Microsoft, OpenAI, and Meta are already utilizing CoreWeave GPUs for training and other AI processes, highlighting the effectiveness and performance improvements they provide compared to prior iterations.
As enterprises advance in their AI journeys, CoreWeave’s infrastructure is poised to accommodate their increasing demands.
