NVIDIA's Blackwell chips are reportedly getting too hot to handle
NVIDIA's latest artificial intelligence (AI) chips, dubbed Blackwell, are suffering from severe overheating problems, as per Reuters. The chips are meant to be crammed into server racks that can hold as many as 72 of them. However, the high temperatures from this setup have sparked concerns about delays in the rollout of NVIDIA's new AI tech.
NVIDIA seeks solutions to overheating problem
The chip overheating issue has made it difficult for NVIDIA to fix. Reportedly, the company has asked suppliers to make a number of changes to the server rack design, in order to tackle the problem. These last-minute modifications have been confirmed by employees, customers, and suppliers speaking to Reuters. However, an NVIDIA spokesperson has played down the overheating issue, claiming that working with cloud service providers and going through several engineering iterations is a part of the process.
Overheating concerns delay Blackwell chip shipment
The Blackwell chips represent a major leap in AI hardware for NVIDIA. Each chip marries two large silicon squares and provides a performance boost, reportedly 30 times faster for AI tasks. The Blackwell chips, unveiled in March, were set for shipment in the second quarter. However, with the overheating concerns, these timelines have started to shift. While NVIDIA has not officially announced delays to its clients, late-stage design changes are a worrying sign for customers like Meta, Google, and Microsoft.
Previous setbacks and future uncertainties
Notably, this isn't the first setback for the Blackwell chips. Earlier development issues had already delayed their release by at least one quarter. The ongoing rack redesigns only add to this uncertainty. NVIDIA has not yet confirmed if the new server designs are finalized, leaving questions about potential impacts on future shipments or further disruptions.
