Are Nvidia’s new AI chips facing overheating snag?
The Blackwell graphics processing units overheat when connected together in server racks designed to hold up to 72 chips, according to a report by the Information.
Nvidia’s Blackwell AI chips are facing problems with accompanying servers that overheat, with some customers worrying they will not have sufficient time to get new data centers up and running, according to a report by the Information.
These new AI chips, which were unveiled in March, have already faced delays. The Blackwell graphics processing units overheat when connected together in server racks designed to hold up to 72 chips, Reuters reported, quoting the Information.
The chips aim to meet the exploding demand for AI across fields and to speed up the training of large language models and other AI systems.
Nvidia has asked its suppliers to change the design of the racks multiple times to resolve the overheating glitch, the source said.
“Nvidia is working with leading cloud service providers as an integral part of our engineering team and process. The engineering iterations are normal and expected,” a spokesperson of the chipmaking giant told Reuters.
Nvidia had earlier said these new Blackwell chips would be shipped in the second quarter, but delays may have impacted customers like Google and Microsoft.
Blackwell chips are designed to significantly enhance the speed and efficiency of AI tasks and can deliver up to 20 petaflops of compute power, which makes them four times faster than Nvidia’s older top chip, according to an earlier report by PCOutlet.
They also reportedly reduce power needs by up to 25 times compared to previous versions.
In October, Nvidia’s rival AMD launched a new AI chip called the Instinct MI325X, which is being pitted against the Blackwell chip.
Gartner has predicted that AI chip revenue will skyrocket in 2024.
