Based on architecture Hopperthe H100 GPU will be the new “monster” of NVIDIA for the world of server And data center. Made up of ben 80 billion transistorsH100 will have an operating frequency of 1.40-1.50 GHz. They will be present 8.448 FP64 And 16,896 FP32 coresas well as 538 Tensor Core, but one of the most notable features will be the use of 96GB of HBM3 memory. However, due to ECC support and other factors, users will be able to only access 80GB across the large 5,120-bit bus.
According to an article posted on the NVIDIA developer page, the H100 was designed with EDA tools (Electronic Design Automation) advanced with the help of Artificial Intelligence through the PrefixRL methodology. This allowed the company to accomplish smaller, faster and more energy efficient chips.
H100 was defined by the same company as the fastest AI chip in the world, with almost 13,000 circuits designed by Artificial Intelligence, which allowed to reduce the area by 25% compared to EDA tools. Since PrefixRL is very computationally demanding, NVIDIA has developed Raptora platform that makes it possible to distribute tasks between CPU, GPU and Spot instances.
Arithmetic circuits were once the craft of human experts, and are now designed by AI in NVIDIA GPUs. H100 chips have nearly 13,000 AI designed circuits! How is this possible? Blog https://t.co/PpKrAmV8vc + a thread pic.twitter.com/3RrZl2muJ3
– Rajarshi Roy (@rjrshr) July 8, 2022
As demonstrated by NVIDIA itself, therefore, the use of Artificial Intelligence for the design of integrated circuits can only improve and speed up operations, so as to have more performing and efficient GPUs in the future. For further information, we refer you to the developer blog.
NVIDIA will begin shipping the Hopper H100 GPU computes during the second half of this year and, although the price has not yet been specified, it will certainly not be an affordable product. Find more details about it in our previous dedicated news.
