NVIDIA’s Top GPUs Are In Big Demand For AI, Places Additional Orders At TSMC
By: Hassan Mujtaba
NVIDIA has placed additional wafer supply orders at TSMC due to the growing demand for its top AI GPUs such as the A100 & H100.
NVIDIA’s AI Success Blows Up AI GPU Demand, Company Has To Put More Orders At TSMC To Retain Supply Balance
In a previous report, we mentioned how the demand for HPC and AI GPUs from NVIDIA can disrupt the supply of gaming chips as the company wants to allocate more resources towards AI which it’s calling a revolution for the PC and tech industry. It was expected that the demand would sooner or later outstrip the supply but NVIDIA is working round the clock to make sure that it can still offer a decent supply of chips to its biggest partners who are willing to pay the extra price of securing the top AI chips in the world. NVIDIA has also been the major driving force behind ChatGPT & several thousands of its GPUs are powering the current and latest models.
Now, DigiTimes has reported that NVIDIA is placing additional orders at TSMC for chips that utilize the CoWoS packaging technology, CoWoS which is otherwise known as Chip-on-Wafer-on-Substrate is a packaging technology that is deployed within NVIDIA’s top data center and cloud GPUs that use HBM memory. Currently, NVIDIA’s Ampere line of A100 & Hopper line of H100 and their derivatives are utilizing this technology and they also happen to be the same chips that are being used for AI & machine learning applications.
Nvidia has recently obtained TSMC’s commitment to CoWoS support for an additional 10,000 wafers in 2023, the sources said, adding TSMC may have to give Nvidia extra CoWoS support for 1,000 to 2,000 wafers monthly throughout the rest of the year.
TMSC has a monthly CoWoS capacity of 8,000 to 9,000 wafers, and extra demand from Nvidia means the foundry’s CoWoS supply will become tight, the sources said.
Nvidia is optimistic about demand for AI chips, but it needs one-stop support from TSMC for both chip manufacturing and advanced packaging, the sources said.
TSMC can churn out around 8000-9000 CoWoS wafers per month but NVIDIA has dropped in an order of an additional 10,000 wafers for the entirety of 2023 which means that TSMC will have to add an additional 1000-2000 wafers throughout the year and significantly tighten its supply. This increased demand has also made TSMC optimistic about the growth of its CoWoS technology.
Bloomberg updated the article
Frank Shaw, a Microsoft spokesman, denied that AMD is part of Athena. “AMD is a great partner,” he said. “However, they are not involved in Athena.”
— Dylan Patel (@dylan522p) May 5, 2023
One more applicant of the CoWoS technology is AMD who will be supplying its Instinct-class chips to Microsoft. There have been reports that Microsoft and AMD are working on a brand-new chip to tackle AMD under the Athena codename however this was debunked and Microsoft is simply going to use existing and upcoming AMD accelerators to power its AI agenda.
There’s also huge demand coming in from China which is still trying to procure NVIDIA’s latest AI GPUs despite the variant being offered to them being cut down & lacking serious interconnect fabrics.
The prices being offered to the Chinese are also much higher than the rest of the world where the US hasn’t applied its technology sanctions. It will be interesting to see if NVIDIA pulls back some of the A800 and H800 GPU production in favor of the standard A100/H100 GPUs. These chips are already using defective dies and since A100/H100 are not entirely perfect, the extra production of standard chips can allow for an increased supply of A800/H800 defective GPUs too.