NVIDIA CEO: Blackwell chip design flaw has been fixed, expected to ship in Q4
NVIDIA CEO Jensen Huang stated that the design flaw of the Blackwell AI chip has been fixed with the help of TSMC and is expected to be shipped in the fourth quarter. The defect had previously affected production, leading to low yield rates. Huang emphasized that there is strong demand in the market for the Blackwell series processors, with Dell and Google already starting to ship and operate related products
According to the Zhitong Finance and Economics APP, Jensen Huang, CEO of NVIDIA (NVDA.US), stated on Wednesday that with the help of TSMC (TSM.US), the design flaw in NVIDIA's latest Blackwell AI chip has been fixed, which had previously affected production. The CEO said, "We have a design flaw in Blackwell, it's functional, but the design flaw resulted in a very low yield. This is 100% NVIDIA's fault."
NVIDIA released the Blackwell chip in March this year, previously stating that it would ship in the second quarter, but it was later delayed, potentially affecting customers such as Meta (META.US), Alphabet (GOOGL.US), and Microsoft (MSFT.US).
Earlier this month, Huang Renxun mentioned that the market demand for the new Blackwell series processors is "crazy." Last week, Dell announced that they will soon start shipping devices with NVIDIA's Blackwell AI accelerator, and Google stated that they have started running servers based on the Blackwell series chips. Previously, OpenAI and Microsoft also made collaboration announcements.
According to previous reports from foreign media, the production delay caused tension between NVIDIA and TSMC, but Huang Renxun believes this is "fake news."
He said, "To make Blackwell work, we designed seven different types of chips from scratch and had to put them into production simultaneously," "What TSMC did was help us recover from production difficulties and restore Blackwell production in an incredible way."
NVIDIA's Blackwell chip combines two silicon wafers of the size of the company's previous products and merges them into a single component, increasing speed by 30 times when performing tasks such as providing answers for chatbots.
At a recent Goldman Sachs conference, the CEO mentioned that these chips will ship in the fourth quarter.
On Wednesday, Huang Renxun unveiled a new supercomputer named Gefion in Denmark, which features 1528 graphics processing units (GPUs) and was built in collaboration with the Novo Nordisk Foundation, the Danish Export and Investment Fund, and NVIDIA