Nvidia CEO Jensen Huang has confirmed that a design flaw that was impacting the expected yields from its Blackwell AI GPUs has been addressed.
Huang clarified that the issue was the result of Nvidia’s design, and not its fabricating partner TSMC, with mass production of the Blackwell B100 and B200 GPUs expected to resume later this month, according to Reuters.
In a conference call in August, Huang revealed Nvidia had modified the design of the Blackwell chips to improve yield, with rumors soon swirling that the incident had caused tension between the company and TMSC.
Speaking at Nvidia’s AI Summit in Mumbai, Huang shed some light on the issue, clarifying that it was the sole result of an Nvidia error, and TSMC was in no way responsible for any production delays.
"We had a design flaw in Blackwell ... It was functional, but the design flaw caused the yield to be low. It was 100% Nvidia's fault,” Huang stated.
Huang pushed back on suggestions that two companies' relationship had become strained, adding that TSMC was instrumental in assisting Nvidia recover from the interruption and get units back in production as soon as possible.
"In order to make a Blackwell computer work, seven different types of chips were designed from scratch and had to be ramped into production at the same time,"he explained
"What TSMC did, was to help us recover from that yield difficulty and resume the manufacturing of Blackwell at an incredible pace."
Huang labels claims of bad blood with TSMC ‘fake news’
After initial reports of the design flaw emerged, some outlets claimed the issue and resulting delays to production may have soured its longstanding relationship with TSMC.
Some outlets suggested the fault was associated with the packaging technology onboard the Blackwell chips, namely TSMC’s CoWoS platform.
According to some reports, the two companies were in deadlock over who was responsible for the fault, blaming each other for the performance issues and potentially jeopardizing the pair’s 30-year relationship.
TSMC has been central in Nvidia’s rise to become the preeminent chipmaker of the AI era, with Huang himself remarking that “without TSMC, there would be no Nvidia today”.
Huang labeled any claims of ill will between the two firms as ‘fake news’.
Speaking at a recent Goldman Sachs conference, the Nvidia CEO announced the Blackwell GPUs will enter mass production in the fourth quarter (late October), and are expected to start shipping in early 2025.
However, Nvidia disclosed that it would have to ship some of the low-yield units in 2024 to satisfy surging demand for the Blackwell GPUs, specialized for AI workloads.
