(Reuters) -Nvidia’s new Blackwell AI chips, which have already confronted delays, have encountered issues with accompanying servers that overheat, inflicting some clients to fret they won’t have sufficient time to get new knowledge facilities up and operating, the Info reported on Sunday.
The Blackwell graphics processing models overheat when linked collectively in server racks designed to carry as much as 72 chips, the report mentioned, citing sources conversant in the difficulty.
The chipmaker has requested its suppliers to vary the design of the racks a number of occasions to resolve overheating issues, in keeping with Nvidia (NASDAQ:) workers who’ve been engaged on the difficulty, in addition to clients and suppliers with data of the difficulty, the report mentioned with out naming the suppliers.
“Nvidia is working with main cloud service suppliers as an integral a part of our engineering crew and course of. The engineering iterations are regular and anticipated,” an organization spokesperson mentioned in an announcement to Reuters.
trH" title="© Reuters. The logo of NVIDIA as seen at its corporate headquarters in Santa Clara, California, in May of 2022. Courtesy NVIDIA/Handout via REUTERS/File Photo" alt="© Reuters. The logo of NVIDIA as seen at its corporate headquarters in Santa Clara, California, in May of 2022. Courtesy NVIDIA/Handout via REUTERS/File Photo" rel="external-image"/>
In March, Nvidia unveiled Blackwell chips and had earlier mentioned they might ship within the second quarter earlier than encountering delays, doubtlessly affecting clients comparable to Meta Platforms (NASDAQ:), Alphabet (NASDAQ:)’s Google and Microsoft (NASDAQ:).
Nvidia’s Blackwell chip takes two squares of silicon the dimensions of the corporate’s earlier providing and binds them right into a single element that’s 30 occasions speedier at duties like offering responses from chatbots.