Oct 26, 2024 13:00:00

NVIDIA CEO Jensen Huang admits that the AI chip design flaws are 100% NVIDIA's fault and TSMC is not responsible, corrected Blackwell chips are in production

When NVIDIA's AI chips 'B100' and 'B200' were first

announced in March 2024 , they were said to be available in late 2024. However, it has since been reported that production has been delayed due to design defects. NVIDIA CEO Jensen Huang admitted that 'NVIDIA is 100% to blame' for the delay in the production of AI chips, and revealed that an improved version of the AI chip is in the works.

Nvidia's design flaw with Blackwell AI chips now fixed, CEO says | Reuters
https://www.reuters.com/technology/artificial-intelligence/nvidias-design-flaw-with-blackwell-ai-chips-now-fixed-ceo-says-2024-10-23/

Nvidia's Jensen Huang admits AI chip design flaw was '100% Nvidia's fault' — TSMC not to blame, now-fixed Blackwell chips are in production | Tom's Hardware

https://www.tomshardware.com/tech-industry/artificial-intelligence/nvidias-jensen-huang-admits-ai-chip-design-flaw-was-100-percent-nvidias-fault-tsmc-not-to-blame-now-fixed-blackwell-chips-are-in-production

The B100 and B200 are the first AI chips to adopt NVIDIA's new GPU architecture, Blackwell , and were announced simultaneously with Blackwell at the technology conference GTC 2024 in March 2024.

NVIDIA announces GPU architecture 'Blackwell' and new GPU 'B200' to realize AI models with trillions of parameters - GIGAZINE

When the announcement was first made, NVIDIA said that the B100 and B200 would be released in late 2024, but it was later reported that design flaws had been discovered in the manufacturing process, delaying shipments of these AI chips.

It is leaked that the production of NVIDIA's new AI chips is expected to be delayed by at least three months - GIGAZINE

According to Tom's Hardware, the B100 and B200 have mismatched thermal expansion characteristics between the GPU chiplets, LSI bridges, RDL interposers, and motherboard substrates, which causes the system to warp and become more susceptible to failure.

Since the reports of the shipping delays were released, some media outlets have reported that the delays are due to NVIDIA's production partner TSMC, and that tensions are growing between NVIDIA and TSMC.

In response to these reports, Huang said, 'There is a design flaw in the chips that use Blackwell, which is resulting in delays in shipments. This is 100% NVIDIA's fault.' He also dismissed the theory that TSMC is to blame for the delays as 'fake news.'

According to Huang, to fix the design flaws, NVIDIA had to redesign seven chips from scratch and start producing the corrected AI chips at the same time. During this process, TSMC helped us resume production of Blackwell-based chips at an incredible pace, Huang said.

Tom's Hardware commented, 'It typically takes about three months to identify, fix, and redesign a chip. The speed at which NVIDIA and TSMC fixed the issues with the B100 and B200 and started reproducing them is incredible.'