NVIDIA CEO Jensen Huang has been on a "critical" trip in recent days, visiting Taiwan to "streamline" the supply of next-generation Hopper H200 and Blackwell B100 GPUs. This visit marks the company’s first step towards the next generation of Hopper H200 and Blackwell B100 artificial intelligence graphics processors, which will reshape the market landscape.
Not only is NVIDIA making leaps and bounds across its AI portfolio, but the company is also working hard to cater to various regional interests despite setbacks like new U.S. regulations. We reported yesterday on Jen-Hsun Huang's visit to Beijing, where he not only participated in festivities related to the Beijing New Year, but also reportedly held meetings with several Chinese customers, including Alibaba and Tencent, to assure them that Nvidia is here to stay.
According to reports, after visiting China, Huang Renxun also decided to go to Taiwan to visit major suppliers of companies such as TSMC and Wistron to learn about the mass production progress of the company's next-generation H200 and B100 AIGPU. This year will mark the transition to a new generation of artificial intelligence products, which is why 2024 will be a "defining" year not only for Nvidia, but also for other companies participating in the artificial intelligence race. It is predicted that global artificial intelligence shipments are expected to grow significantly this year and are expected to reach about 350,000 pieces, of which NVIDIA will occupy a considerable share.
Regarding NVIDIA's upcoming products, the first is its Hopper H200 GPU, which is equipped with Micron's HBM3e solution, with a memory capacity of 141GB and a bandwidth of up to 4.8TB/s. Compared with the NVIDIA A100, the bandwidth has increased by 2.4 times and the capacity has doubled.
In applications such as Llama2 (70 billion parameter LLM), this new memory solution nearly doubles NVIDIA's AI inference performance compared to the H100 GPU. The next blockbuster product is the Blackwell B100AIGPU. In terms of performance, it will bring a decisive improvement in performance per watt through the use of chipset design.
NVIDIA's next-generation Hopper H200 and Blackwell B100 artificial intelligence graphics processors are expected to debut within 2024. It is reported that due to the high adoption rate of Hopper in 2024, the B100 will be delayed to the fourth quarter. This is going to be an interesting year for the tech industry as a whole, especially with the advancements in artificial intelligence and the markets associated with it. Nvidia is expected to announce more information about artificial intelligence and data center chips at the GTC 2024 conference in March.