Nvidia’s next-generation Blackwell data center processors are encountering serious overheating issues, particularly when installed in high-capacity server racks. This predicament has prompted necessary redesigns of the racks, which are expected to delay product shipping and raise concerns among major clients such as Google, Meta, and Microsoft regarding their deployment timelines.
According to a report from The Information, Blackwell GPUs are struggling to cope with the heat generated in ultra-dense servers that accommodate 72 processors. Each Blackwell processor consumes over 1000 watts, leading to excessive heat accumulation in a compact environment.
In response to these challenges, Nvidia is collaborating with its suppliers and partners to implement revisions and design changes aimed at mitigating the overheating problems. Although such redesigns are not unprecedented, the current situation is pushing back the anticipated shipping schedule that was originally slated for this quarter.
This isn’t the first time Blackwell has faced scrutiny; earlier reports suggested that Nvidia and its manufacturing partner, TSMC, were grappling with yield issues due to the processor’s packaging design. However, those concerns were seemingly resolved, as indicated during Nvidia’s quarterly earnings call.
As Nvidia prepares to report earnings on November 20, a spokesperson has commented, pointing out that the GB200 systems are the most advanced computers ever devised. The integration into diverse data center environments requires close collaboration with clients, and the engineering adaptations being made align with expectations.
However, Anshel Sag, a principal analyst at Moor Insights & Strategies, remains skeptical of the situation, questioning whether the overheating issue is broadly affecting the chips or if it pertains to specific configurations. He noted the timing of the report is curious, considering the Supercomputing 24 conference is underway. Sag implied that the release of such rumors may serve competitive interests, suggesting a potential strategy to shift attention from Nvidia to rival platforms during a critical industry event.
