English
Français
Deutsch
Italiano
Português
Русский
Español
简体中文
日本語
Generative artificial intelligence applications will run better starting next year.
On Tuesday,
Nvidia
(ticker: NVDA) announced the next-generation version of its GH200 Grace Hopper Superchip. The update will incorporate the world’s first use of HBM3e advanced memory technology, offering 1.7 times the memory capacity and 1.5 times the memory bandwidth of
Nvidia’s
current top-of-the-line H100 chip.
“We’re giving Grace Hopper.. a boost with the world’s fastest memory,” CEO Jensen Huang said during his SIGGRAPH keynote speech in Los Angeles. “This processor is designed for a scaleout of the world’s data centers.”
So-called superchips link together Nvidia’s Grace central processing unit, or CPU, and its Hopper GPU, or graphics processing unit, so that they can work more efficiently together. The new version will have 141 gigabytes of HBM3e memory with 5 terabytes per second of bandwidth.
The company didn’t disclose a price. The new chips are expected to be available in the second quarter next year.
The new GH200’s memory is “much faster,” Nvidia executive Ian Buck said during a videoconference call with reporters Tuesday. The larger memory capacity also allows for bigger AI models to run on a single GPU, offering better performance in certain instances, he said.
The chip was designed to handle “the world’s most complex generative AI workloads, spanning large language models, recommender systems and vector databases,” Nvidia said in a release.
Nvidia dominates the market for chips used for AI applications. Its products are widely used for generative AI, which has been trending since OpenAI’s ChatGPT was released late last year. The technology ingests text, images, and videos and then uses the patterns it finds to create content.
Last month,
Microsoft
(MSFT) and Google’s parent
Alphabet
(GOOG) both said they plan to invest aggressively in AI infrastructure over the next year.
Write to Tae Kim at [email protected]
Read the full article here
