Nvidia Blackwell B200 Chip Delays? Or Not Delayed?

by

There have been reports that Nvidia was delaying its next-generation artificial intelligence chips by at least three months. Mass shipments may not take place until early 2025.

There are other reports that to maximize the available CoWoS-L capacity, Nvidia is replacing B100 with B200A which is using CoWoS-S, thereby freeing up scarce CoWoS-L capacity for B200/GB200 – this is higher value for Nvidia and also better matches customer demand mix as the vast majority of customers want rack-scale GB200 but most (other than two) are not fully ready with their liquid cooling infrastructure and the new B200A is allowing the company to offer a new 64 rack option (GB200A-NVL64) that is air cooled, according to the analysts.

The rumors regarding the delay of Nvidia’s Blackwell B100 chips appear credible, based on information available up to August 18, 2024. Here’s a summary:

Multiple Sources: Various reports and posts on X from credible tech news outlets and industry insiders have confirmed that Nvidia has encountered design flaws with its Blackwell series, particularly affecting the B100 chip. These sources mention delays ranging from three months to pushing the release into 2025.
Nvidia’s Official Stance: While Nvidia has not directly confirmed the delay in public statements, they’ve mentioned being “on track” for production in the second half of the year, which could be interpreted as an acknowledgment of some delay or adjustment in their original timeline.
Customer Notifications: Reports indicate that Nvidia has informed major customers like Microsoft and Meta about these delays, suggesting an official communication regarding the setback.
Technical Issues: The delay is attributed to design flaws, specifically involving the complex chip-on-wafer-on-substrate (CoWoS) packaging technology from TSMC, which is crucial for the Blackwell architecture’s performance.
Market Impact: This delay could affect Nvidia’s market position, especially in the AI computing space, where competitors might see an opportunity to gain ground.

Given this information, while Nvidia might be working to mitigate the impact of these delays, the consensus from industry reports supports the notion that there indeed are delays with the Blackwell B100 chips. However, Nvidia might still find ways to expedite certain aspects of their rollout. Nvidia share prices have pretty much fully recovered.

3 thoughts on “Nvidia Blackwell B200 Chip Delays? Or Not Delayed?”

  2. Blackwell is dead on arrival. A warping overheat of the communication bridge chip underneath two monolithic GPU’s is not curable until Nvidia designs a chip to chip edge connection. Designing a new interconnect technology is a three to five year research project.

    Nvidia’s marketing advantage is proprietary easy to use CUDA software for AI training. AMD checkmated CUDA by purchasing software and writing an open source alternate to CUDA. For ease of use AMD purchased 100 software engineers who can deliver solutions to the IT managers and system builders.

  3. Nvidia value relies in a specialized and improved branch of his GPU technology that was developed in the past to work on another domains (graphics an general calculations) that shares a few characteristics with AI needs, but it is not 100% taylor made for that, and it isn´t a real breackthrought technology for AI.
    It is probable that a new generation of some sort of ASICs will overshoot nvidia technology, leaving nVidia in a no comfortable situation in the mid term. Nvidia will have many problems to further develp it´s “classic” technology. It is probably a dead end.

Comments are closed.