The Blackwell graphics processing items overheat when linked collectively within the personalized server racks the corporate has designed, the report stated, citing sources accustomed to the problem.
The AI chipmaker has requested its suppliers to vary the design of the racks a number of instances to resolve overheating issues, in accordance with Nvidia workers who’ve been engaged on the problem, in addition to clients and suppliers with data of it, the report stated with out naming the suppliers.
Nvidia didn’t instantly reply to a request for remark exterior common enterprise hours.
Nvidia unveiled Blackwell chips in March and had earlier stated they’d ship within the second quarter earlier than encountering delays, doubtlessly affecting clients akin to Meta Platforms , Alphabet’s Google and Microsoft.
Nvidia’s Blackwell chip takes two squares of silicon the scale of the corporate’s earlier providing and binds them right into a single part that’s 30 instances speedier at duties like offering responses from chatbots.