Nvidia "exploded" late at night, and AI chips swept the technology circle
Nvidia, a world-renowned chip manufacturer, released the world's strongest AI chip H200 at the 2023 Global Supercomputing Conference (SC23) on the 13th, which has increased performance by about 60% to 90% compared with its predecessor H100, which has attracted great attention from the industry.
(Note: H200 is an upgraded version of Nvidia H100, both are also based on the Hopper architecture, the main upgrade of H200 includes 141GB of HBM3e video memory, and the memory bandwidth has been increased from 3.35TB/s to 4.8TB/s of H100.)
The American technology website "techwireasia" said on the 13th that Nvidia is using its most powerful chip to date, H200, to promote the artificial intelligence revolution. It's been almost a year since OpenAI launched ChatGPT, and the global demand for AI chips is stronger than ever. Today, most big tech companies are focusing their attention on generative AI. Nvidia produces the largest number of GPUs with the highest performance. After releasing dozens of chips to cater to the exponentially growing AI market, the graphics chip giant has launched its most powerful GPU yet – the H200. The H200 launch comes as Nvidia struggles to defend its dominance in AI computing, with a slew of chip startups and cloud service providers such as Intel, Chaowei Semiconductor, and Amazon Web Services trying to capture more market share amid surging demand for chips driven by AI workloads. Major players such as AWS, Google, Microsoft, and traditional AI and high-performance computing entities such as Chaowei Semiconductor are actively preparing for the next generation of processors for training and inference.
Some analysts believe that this chip will be used in a new generation of supercomputers. However, the website of the American technology media "The Verge" believes that the big question is whether the parties will be able to get the new chips, or whether they will be limited in supply like the H100 - Nvidia does not have a definite answer to this.
The first batch of H200 chips will be available in the second quarter of 2024, and Nvidia said it is working with "global system manufacturers and cloud service providers" to deliver the chips. At the same time, the chip could be expensive, although Nvidia does not currently list the price of the new chip. According to a report on the CNBC website in the United States on the 13th, H100 is estimated to cost between $25,000 and $40,000 each, and thousands of H100 chips may be needed to operate at the highest level.
