Jeff Clarke, chief operating officer and vice chairman of Dell, has announced that his company had started shipments of servers based on Nvidia's Blackwell GPUs for AI and HPC and Grace CPUs. The machines use liquid-cooled PowerEdge XE9712 racks and are aimed at enterprises.
"Incredibly excited to ship the industry's first enterprise-ready Nvidia GB200 NVL72 server racks," Clarke wrote in an X post. "These fully integrated, liquid-cooled PowerEdge XE9712 racks are a huge milestone, the foundation for Dell AI Factory with Nvidia to advance HPC and AI innovation."
Dell mentioned CoreWeave, a cloud service provider focused on AI that has very close ties to Nvidia. As the company exclusively uses Nvidia GPUs, it gets preferential treatment and product allocation from the green company, so it is not particularly surprising that CoreWeave is among the first CSPs to get Nvidia's Blackwell hardware.
"We are proud to bring up the first Nvidia GB200 NVL72 from Dell with Nvidia Quantum InfiniBand, setting a new bar for AI infrastructure," an X post by CoreWeave reads. "This would not have been possible without the support of our valued partners at Dell and Switch."
It should be noted that the GB200 hardware that Nvidia ships today is based on the initial low-yield revision of B200 processors. These GPUs are in relatively short supply, so do not expect the company to ship a boatload of such processors. Officially, the company expects to ship several billion dollars worth of Blackwell hardware this fiscal year, which is significantly below the company's earnings from Hopper hardware in the most recent quarter.
Yet, it is strategically important for Nvidia to start shipments of Blackwell GPUs rather sooner than later in a bid to enable its partners to tailor their software for the latest architecture.
Each Nvidia NVL72 server rack is projected to consume up to 120kW of power (up from 40kW in the case of Hopper-based racks), which is significantly higher than typical server racks currently used by cloud server providers. To install such machines into their datacenters, CSPs may have to redesign their server rooms in a bid to feed such servers with sufficient power.
