Computing power for generative artificial intelligence applications will soon increase significantly.
On Monday, Nvidia (ticker: NVDA) announced its new DGX GH200 AI supercomputer with 256 GH200 Grace Hopper superchips. There are a lot of letters and numbers, but the specs are the most important.
The new DGX system will enable the next generation of generative AI applications thanks to its larger memory size and larger model capabilities, Nvidia executives said during a video conference with reporters on Friday. The DGX GH200 will have nearly 500 times more memory than Nvidia’s DGX A100 system.
“The DGX GH200 AI supercomputer integrates NVIDIA’s leading accelerated computing and networking technologies to expand the frontiers of AI,” said NVIDIA CEO Jensen Huang at the COMPUTEX keynote in Taiwan. said.
Nvidia chips have been exposed to this trend of generative AI since OpenAI released ChatGPT late last year. This technology brute-forces ingesting text, images, and video to create content.
Ads – scroll to continue
Chatbots like ChatGPT generate human-like responses, or best guesses, based on word relationships found by digesting previously written content on the internet or from other forms of text. Use language models.
Nvidia hopes the new supercomputer will enable developers to build better language models and complex recommendation algorithms for AI chatbots to create more effective fraud detection and data analysis.
The DGX GH200 incorporates 256 GH200 superchips. Each superchip is equipped with a GPU. So the DGX GH200 has 256 GPUs, while the previous model had 8.
There are two types of logic chips: the central processing unit (CPU), which serves as the main computational brain of a PC/server, and the graphics processing unit (GPU), which is used for gaming and AI calculations.
Nvidia mentioned Alphabet’s (GOOGL) Google Cloud, Meta (META), and Microsoft.
Ads – scroll to continue
(MSFT) will be one of the first companies to access the DGX GH200 to explore the capabilities of generative AI. The system is expected to be available by the end of 2023, but the company has not disclosed pricing.
Nvidia also announced that the GH200 Grace Hopper superchip has entered full production. Superchip uses NVLink connectivity technology to link his Nvidia’s Grace CPU and Hopper GPU.
Nvidia shares soared 24% the day after Thursday after the company announced earnings forecasts for the quarter that were significantly better than Wall Street expected. Company management said the rise was driven by substantial demand for its AI data center products from cloud computing providers, large consumer internet companies, start-ups and others.
Email Tae Kim (tae.kim@barrons.com).
