Delay in the launch of the next NVIDIA Blackwell GPUs: explanations

According InformationNVIDIA is facing a design flaw in its upcoming Blackwell GPUs for consumer and data centers. The problem could lead to a three-month delay in the launch of these chips.

NVIDIA had originally announced the B100 and B200 GPUs, based on the Blackwell architecture, for March 2024. On July 29, the company confirmed that it had shipped test samples. Everything seemed to be going according to plan.

However, leaks have revealed issues with the dual-GPU versions of these new chips. Specifically, these versions include the B200, a package combining two B100s, and the GB200, a dual-GPU version similar to the B200 but also incorporating a CPU.

A fault affects the part that connects the two GPUs, preventing them from being merged into a single chip. The exact nature of the problem remains unknown. What is certain is that the delay seems inevitable. NVIDIA has already informed its main customers of the difficulties they have encountered with these dual-GPU chips.

Will the RTX 5000 also be delayed?

This raises a crucial question: will this delay also affect… The launch of the RTX 5000 graphics cards? Currently, chips for data centers and enterprises, especially those used for artificial intelligence, have become NVIDIA's main source of revenue. This phenomenon has been further amplified by this year's AI boom.

In comparison, consumer graphics cards like the RTX are making less and less money, so it's likely that TSMC's production lines will prioritize these professional chips.

It's also worth remembering that the RTX 5000s use TSMC's same N4P manufacturing process as its server chips, reinforcing the hypothesis that a delay could also affect gaming graphics cards.