"100% Nvidia's fault": Jensen Huang admits Blackwell AI chips had a concerning design flaw

Nvidia CEO Jensen Huang has confirmed a creation flaw successful its top-end Blackwell AI chips which had affected accumulation was an wholly soul problem, which has now been fixed.

"We had a creation flaw successful Blackwell," Huang said astatine an arena successful Copenhagen, Reuters reported. "It was functional, but nan creation flaw caused nan output to beryllium low. It was 100% Nvidia's fault."

First identified successful August 2024, nan hold to Blackwell B100/B200 processors had raised eyebrows astir nan world, but Huang reassured that it was Nvidia's ain doing that caused nan issue.

Blackwell delays

Blackwell chips person been successful precocious request since Nvidia unveiled nan level earlier successful 2024, pinch Huang describing it as, "the world's astir powerful chip," offering antecedently unheard-of levels of AI computing power.

Set to statesman shipping successful nan second portion of 2024, Blackwell binds together 2 GPU dies, which are connected by 10 TB/second chip-to-chip nexus into a single, unified GPU. This uses utilizing TSMC's CoWoS-L packaging technology, which relies connected an RDL interposer equipped pinch section silicon interconnect (LSI) bridges that request to beryllium located specifically to let accelerated information transportation - nan misalignment of which resulted successful nan issue.

Initial media reports had claimed nan rumor had caused clash pinch manufacturing partner TSMC, but Huang dismissed nan claims arsenic "fake news".

"In bid to make a Blackwell machine work, 7 different types of chips were designed from scratch and had to beryllium ramped into accumulation astatine nan aforesaid time," he said.

"What TSMC did, was to thief america retrieve from that output trouble and resume nan manufacturing of Blackwell astatine an unthinkable pace."

Blackwell is group to beryllium up to 30x faster than its Grace Hopper predecessor erstwhile it comes to AI conclusion tasks, whilst besides reducing costs and power depletion by up to 25x.

More from TechRadar Pro

• Lenovo CEO - get fresh for a caller era of hybrid AI for all
• Nvidia CEO - AI could beryllium nan largest technological leap we’ve ever seen
• These are nan best mobile workstations astir today