Nvidia CEO Jensen Huang has confirmed that a design flaw in its top-end Blackwell AI chips, which affected production, was a completely internal problem, which has now been resolved.
“We had a design flaw in Blackwell,” Huang said at an event in Copenhagen, Reuters reported. “It was functional, but the design flaw caused the yield to be low. It was 100% Nvidia’s fault.”
The delay in Blackwell B100/B200 processors, which was first discovered in August 2024, had raised eyebrows around the world, but Huang assured that the problem was Nvidia’s own doing.
Blackwell delays
Blackwell chips have been in high demand since Nvidia unveiled the platform earlier in 2024, with Huang describing it as “the most powerful chip in the world,” offering previously unheard levels of AI computing power.
Expected to hit the market in late 2024, Blackwell combines two GPU chips connected via a 10TB/second chip-to-chip link into a single, unified GPU. This uses TSMC’s CoWoS-L packaging technology, which relies on an RDL interposer equipped with local silicon interconnect (LSI) bridges that must be specifically located to enable high-speed data transfer – of which the wrong alignment caused the problem.
Initial media reports claimed the issue had caused friction with manufacturing partner TSMC, but Huang dismissed the claims as “fake news”.
“To make a Blackwell computer work, seven different types of chips were designed from scratch and had to be put into production at the same time,” he said.
“What TSMC did was help us recover from those yield issues and resume Blackwell production at an incredible pace.”
Blackwell will be up to 30x faster than its Grace Hopper predecessor when it comes to AI inference tasks, while also reducing costs and energy consumption by up to 25x.