Nvidia’s latest Blackwell AI chips are allegedly overheating in test runs with associated server racks. According to a report by The Information, which broke the story, the heat management problems with the 72-chip racks are leading data centre providers to worry that their rollout of Nvidia’s new chips will not meet internal deadlines. Nvidia has asked its suppliers to make multiple design modifications multiple times to address the problem, added the publication, citing staff at the GPU design firm who have been working on the matter.
Nvidia’s Blackwell chips were introduced in March, featuring a dual-silicon architecture that significantly increases processing speeds, making them 30 times faster than earlier models for tasks such as chatbot responses. The chip incorporates six new technologies, enabling AI training and real-time large language model (LLM) inference for models with up to 10 trillion parameters.
Nvidia’s Blackwell AI chip
Initially scheduled for release in the second quarter, the Blackwell chips have already faced delays due to earlier design flaws, which Nvidia claims have since been resolved.
Despite the overheating challenges, Nvidia has not formally notified customers of further delays. The delays and technical setbacks have raised concerns among Nvidia’s key clients. Microsoft, Meta Platforms, and Alphabet’s Google are among those awaiting delivery of the new server racks, which are integral to advancing their AI capabilities.
A spokesperson for Nvidia addressed the challenges in a statement to Reuters. “Nvidia is working with leading cloud service providers as an integral part of our engineering team and process,” they said. “The engineering iterations are normal and expected.”
Last week, Nvidia announced that SoftBank’s Japanese telecom division would be the first to receive chips based on the Blackwell design. Speaking at the Nvidia AI Summit in Japan, CEO Jensen Huang revealed that SoftBank is constructing Japan’s most powerful AI supercomputer using the Blackwell platform. Plans also include adopting the Nvidia Grace Blackwell platform for upcoming systems.
This move is part of a larger partnership between the two companies aimed at advancing AI and reshaping Japan’s telecom sector.
Besides, Nvidia’s shares hit an all-time high in October, edging the Santa Clara-based chipmaker closer to overtaking Apple as the world’s most valuable company. The stock climbed 2.4% to close at $138.07, fuelled by robust investor confidence in the surging demand for Nvidia’s AI processors.