Nvidia’s New AI Supercomputer Is a Game Changer

Text size

Nvidia’s latest AI supercomputer is powered by 256 of these Grace Hopper chips.


Courtesy of Nvidia

Generative artificial intelligence applications will soon receive a massive boost in computing power.

On Monday,

Nvidia

(ticker: NVDA) announced its new DGX GH200 AI supercomputer powered by 256 GH200 Grace Hopper Superchips. That’s a lot of letters and numbers but the specs are what matter most.

The new DGX system will enable the next generation of generative AI applications thanks to its bigger memory size and larger-scale model capabilities, an

Nvidia

executive said during a videoconference call with reporters Friday. The DGX GH200 will have nearly 500 times the memory of Nvidia’s DGX A100 system.

“DGX GH200 AI supercomputers integrate NVIDIA’s most advanced accelerated computing and networking technologies to expand the frontier of AI,” Nvidia CEO Jensen Huang said during his COMPUTEX keynote speech in Taiwan.

Nvidia chips have high exposure to generative AI, which has been trending since OpenAI’s release of ChatGPT late last year. The technology ingests text, images, and videos in a brute-force manner to create content.

Chatbots like ChatGPT use a language model that generates human-like responses, or their best guesses, based on word relationships found by digesting what’s previously been written on the internet or from other forms of text.

Nvidia expects its new supercomputer will allow developers to build better language models for AI chatbots, complex recommendation algorithms, and create more effective fraud detection and data analyses.

The DGX GH200 incorporates 256 GH200 Superchips. Each Superchip has a GPU, which means the DGX GH200 will have 256 GPUs compared to 8 GPUs in the prior model.

There are two types of logic chips: central processing units (CPUs) that act as the main computing brains for PCs/servers and graphics processing units (GPUs) that are used for gaming and AI calculations.

Nvidia said Alphabet’s (GOOGL) Google Cloud,

Meta

(META), and

Microsoft

(MSFT) will be among the first companies to get access to the DGX GH200 to explore its capabilities for generative AI. The system is expected to be available by the end of 2023. The company did not provide a price.

Nvidia also announced the GH200 Grace Hopper Superchip had entered full production. The Superchip links together Nvidia’s Grace CPU and Hopper GPU using its NVLink connecting technology.

On Thursday, Nvidia shares soared 24% a day after the company provided a revenue forecast for the current quarter that was markedly above Wall Street expectations. The company’s management said the upside came from substantial demand for its AI data center products from cloud computing providers, large consumer internet companies, start-ups, and other enterprises.

Write to Tae Kim at [email protected]

Source: https://www.barrons.com/articles/nvidia-ai-supercomputer-a9dff101?siteid=yhoof2&yptr=yahoo