Nvidia's Blackwell processors overheating problem has been revealed
14:18, 20.11.2024
Customers have faced unpleasant consequences after testing the new Blackwell B200 processors, the problem lies in the overheating of servers. At this point, there are real risks that customers will not be able to run their data centers.
What is the actual cause of overheating?
There is a strong possibility that the cooling system is not doing its job due to the use of 72-chip server racks.
According to data from Nvidia, they have already asked their partners several times to support a change in the design of the racks. And also the company has stated that such changes are not unexpected and they were quite predictable. The change in the design may affect the delay in the delivery of accelerators.
These chips were introduced in March of this year and are mainly used for AI-related tasks. Blackwell B200 has 208 billion transistors and a pair of these chips will become the core of GB200. The company has also released the GB200 NVL72 server rack with 72 graphics accelerators.
As for the price, Blackwell B200 will be much cheaper than H100 and is evaluated in the range of $30 to $40 thousand.