Nvidia's Blackwell AI Chip Has Overheating Problems on Servers

According to inside sources, the Blackwell GPU experienced excessive heat when operating in a server cabinet capable of holding up to 72 chips.

Nvidia has asked suppliers to change the structure of server cabinets multiple times to fix the thermal issue, based on feedback from Nvidia engineers and customers with knowledge of the issue provided to The Information.

An Nvidia spokesperson said in a statement that the company is working closely with major cloud providers and their engineering teams to address the issue. Nvidia said these technical adjustments are normal and planned.

In March, Nvidia introduced its Blackwell chip line, which was expected to launch in the second quarter of 2024. However, the plan has been delayed, affecting major customers such as Meta Platforms (Facebook), Alphabet's Google and Microsoft.

The Blackwell chip is intended to be a pioneer in graphics processing and artificial intelligence. With a design consisting of two interconnected silicon cells, this chip line is claimed by Nvidia to be able to improve processing performance up to 30 times compared to the previous generation, especially in applications such as chatbots. The product is expected to play an important role in large data centers and AI applications that require high computing power.

Thanh Mai (t/h)