New Nvidia chips are delayed due to a design flaw that could impact training of next-generation AI models
Nvidia’s new Blackwell AI chips have reportedly been found to have a design flaw, delaying their release. This delay could impact the training of next-generation AI models for major tech companies.
Nvidia’s latest series of artificial intelligence (AI) chips, known as the Blackwell series, is facing launch delays due to design flaws. The Information reports that the problem was caught unexpectedly late in the production process, which could delay the chip’s release by three months or more, affecting the company’s major customers including Meta Platform Inc., Google LLC and Microsoft Corp., which have collectively ordered billions of dollars’ worth of these new chips.
Nvidia’s new Blackwell chips were unveiled earlier this year in March, replacing the Grace Hopper superchip, the company’s current flagship chips. However, official availability is being delayed, as first reported by The Information citing people who help build chips and server hardware for Nvidia. The publication reported that the new chips will not be available for mass shipment until the first quarter of 2025, unlike the original schedule that anticipated a ramp-up in the latter half of 2024.
In response to The Information’s report, an Nvidia spokesperson said in an emailed statement, “As we’ve stated previously, demand for Hopper is strong, broad sampling of Blackwell has begun, and production is on track to ramp up in the second half.”
Nvidia has reportedly informed its IG Tech customers, including Microsoft, about the delay, according to a Microsoft employee and another source familiar with the situation.
Notably, Nvidia has high hopes from its upcoming Blackwell series. While the previous AI chip, the H100, has been highly successful and is currently an integral part of many AI cloud infrastructures, the success of the Blackwell series is more important for the company to maintain Nvidia’s market dominance.
Introducing the new series earlier this year, Nvidia explained that its Blackwell platform represents a major advancement in computing technology, delivering improved performance for real-time generative AI tasks on large language models. The company explained that the Blackwell GPU architecture includes six key technologies that provide 25 times more efficiency in cost and energy consumption than previous models.
This architecture is designed to support a wide range of computational tasks such as data processing, engineering simulation, electronic design automation, and quantum computing. Blackwell GPUs with 208 billion transistors, built using a custom 4NP TSMC process, are capable of supporting AI models with up to 10 trillion parameters.
“Generative AI is the defining technology of our time. Blackwell is the engine powering this new industrial revolution. Working with the world’s most dynamic companies, we will realize the possibilities of AI for every industry,” said Jensen Huang, NVIDIA founder and CEO.
