Official announcements from Nvidia or credible industry reports

Nvidia's Blackwell GPUs Overheat in 72-GPU Servers; Dell Ships First Racks Despite Issues

Nov 18, 2024, 12:02 AM

Nvidia's upcoming Blackwell GPUs, designed for artificial intelligence applications, are reportedly encountering overheating issues in servers, raising concerns among customers about potential delays in data center deployments. According to reports from The Information, the overheating occurs when the GPUs are connected together in custom server racks designed to hold up to 72 chips and weighing 1.5 tons. Nvidia has asked suppliers to redesign the racks multiple times to address the issue, and some customers have canceled or put orders on hold until the problem is resolved. Despite the challenges, Dell has announced it is shipping the industry's first GB200 NVL72 server racks, the PowerEdge XE9712, to customer CoreWeave. Nvidia is working closely with suppliers to resolve the overheating problems, stating that such engineering iterations are normal and expected. News of the overheating issue has impacted Nvidia's stock, which fell by over 2% in pre-market trading.