In a recent interview, NVIDIA's CEO Jensen Huang admitted that the design flaws in the Blackwell AI portfolio were entirely the company's fault. Contrary to earlier rumors, the Taiwan Semiconductor Manufacturing Company (TSMC) had no role in these issues. The problem stemmed from NVIDIA's need to produce seven different chips simultaneously for Blackwell to work. TSMC did, however, help NVIDIA recover quickly from the low yield rates. Now, the first Blackwell products are set to ship by the end of 2024, making it one of the most anticipated releases in NVIDIA's history.

The Blackwell AI portfolio is a high-demand product in the industry due to its performance and capabilities. A few weeks before its launch, rumors circulated about design flaws linked to TSMC's Chip-on-Wafer-on-Substrate (CoWoS) technology. However, Jensen Huang clarified that TSMC was not involved in these problems. The fault lay solely with NVIDIA's design process.

Despite the initial hiccups, NVIDIA aims for Blackwell to be their most successful product to date. The company had to design and produce seven different chips at the same time, leading to lower yield rates. TSMC's assistance in speedily resolving these issues was crucial. As the first Blackwell products prepare for shipping, the tech world eagerly awaits seeing how this new architecture performs.

NVIDIA CEO: Blackwell's Flaws Were Ours, TSMC Helped Fix It

Actions