NVIDIA has launched its next generation of ai supercomputer chips that will likely play an important role in future advances in deep learning and large language models (LLM) like OpenAI’s GPT-4, the company announced. The technology represents a significant leap from the last generation and is ready to be used in data centers and supercomputers, working on tasks such as weather and climate prediction, drug discovery, quantum computing and more.
Key product is NVIDIA-based HGX H200 GPU "Hopper" architecture, a replacement for the popular H100 GPU. It is the company’s first chip to use HBM3e memory, which is faster and has more capacity, making it more suitable for large language models. "With HBM3e, the NVIDIA H200 offers 141 GB of memory at 4.8 terabytes per second, almost double the capacity and 2.4 times the bandwidth compared to its predecessor, the NVIDIA A100." the company wrote.
In terms of ai benefits, NVIDIA says the HGX H200 doubles the inference speed on Llama 2, an LLM of 70 billion parameters, compared to the H100. It will be available in 4-way and 8-way configurations that are compatible with both the software and hardware of H100 systems. It will work in all types of data centers (on-premises, cloud, hybrid cloud, and edge) and will be deployed by Amazon Web Services, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure, among others. It is scheduled to arrive in the second quarter of 2024.
NVIDIA’s other key product is the GH200 Grace Hopper "superchip" which combines the HGX H200 GPU and the Arm-based NVIDIA Grace CPU using the company’s NVLink-C2C link. It is designed so that supercomputers allow "Scientists and researchers to tackle the world’s most challenging problems by accelerating complex ai and HPC applications running terabytes of data." NVIDIA wrote.
The GH200 will be used in "More than 40 ai supercomputers at global research centers, system manufacturers and cloud providers." said company, including Dell, Eviden, Hewlett Packard Enterprise (HPE), Lenovo, QCT and Supermicro. Among them stands out the HPE Cray EX2500 supercomputer that will use four GH200s, expanding up to tens of thousands of Grace Hopper Superchip nodes.
Perhaps the largest Grace Hopper supercomputer is JUPITER, located at the Jülich facility in Germany, which will become the "The most powerful artificial intelligence system in the world." when installed in 2024. It uses a liquid-cooled architecture, "with a boost module comprising nearly 24,000 NVIDIA GH200 Superchips interconnected with the NVIDIA Quantum-2 InfiniBand networking platform," according to Nvidia.
NVIDIA says JUPITER will help contribute to scientific advances in a number of areas, including climate and weather prediction, generating high-resolution climate and weather simulations with interactive visualization. It will also be used for drug discovery, quantum computing and industrial engineering. Many of these areas use custom NVIDIA software solutions that make development easier but also make supercomputing groups dependent on NVIDIA hardware.
New technologies will be key for NVIDIA, which now derives most of its revenue from the artificial intelligence and data center segments. Last quarter, the company posted a record $10.32 billion in revenue in that area alone (out of total revenue of $13.51 billion), up 171 percent from a year ago. He certainly hopes the new GPU and superchip will help continue that trend. Last week, NVIDIA broke its own record for ai training using older H100 technology, so its new technology should help it widen that lead over its rivals in a sector it already dominates.
This article originally appeared on Engadget at https://www.engadget.com/nvidia-announces-its-next-generation-of-ai-supercomputer-chips-140004095.html?src=rss
