Dell on Thursday said that it had delivered industry’s first systems built on Nvidia’s GB300 NVL72 platform to CoreWeave, a leading AI cloud service provider (CSP). CoreWeave said that it had deployed the Blackwell Ultra-based cluster with its data center hosting provider Switch.
Dell’s and CoreWeave’s initial rollout involves Dell Integrated Racks equipped with 72 Nvidia Blackwell Ultra GPUs, 36 Arm-based 72-core Grace CPUs, and 36 BlueField DPUs per rack. Each GB300 NVL72 rack delivers 1.1 ExaFLOPS of dense FP4 inference and 0.36 ExaFLOPS of FP8 training performance, which is 50% higher compared to a GB200 NVL. In addition, each GB300 NVL72 rack features 20 TB of HBM3E and 40 TB of total RAM. Finally, the new rack-scale solution relies on Nvidia’s Quantum-X800 InfiniBand switches and ConnectX-8 SuperNICs for scale-out connections at up to 14.4 GB/s, twice as fast than previously.
These densely configured systems built for maximized training and inference performance are liquid cooled due to their extreme power consumption (up to 1,400W per GPU).
“Dell’s delivery of Nvidia GB300-powered solutions is more than a milestone,” a statement from the company reads. “It reflects the trust our customers and partners continue to place in our expertise. By seamlessly engineering the compute, the network and the storage under one roof and fine-tuning with integration and deployment services, we help our customers move at unprecedented speed and scale. It is the continued innovation and speed that only we can execute that is allowing us to empower incredibly cool customers and accelerate work with partners like CoreWeave, Nvidia and many others.”
Dell noted that the rack-scale systems come assembled and tested as well as are designed for fast installation and deployment for production use. This is the second time when Dell and CoreWeave deploy Nvidia’s latest AI GPU platform ahead of the industry. This achievement results from close work with Nvidia combining hardware, software, and services into integrated offerings that can be rapidly deployed.
The deployment will enable CoreWeave to enhance its cloud platform to handle more advanced large language model training, reasoning processes, and inferencing. As CoreWeave employs more GB300 NVL72-based racks, total performance available to its customers will increase.
One interesting thing to note is that Dell and CoreWeave deploy GB300 NVL72 racks just about seven months after they deployed the first GB200 NVL72 machines, which raises a question about longevity of the original GB200 platform (which was of delayed by a bit, but still). From CSPs point of view, it makes more sense to invest in higher-performance Blackwell Ultra machines than in the original Blackwell systems. So perhaps, there is quite some pent-up demand for Nvidia’s Blackwell Ultra, so Nvidia will enjoy even more sales than it did in the first half of the year.
Follow Tom’s Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.