The NVIDIA H200 NVL
The NVIDIA H200 NVL allows for 1.3x quicker high-performance compute and 1.7x faster large language model inference, making it perfect for lower-power, air-cooled corporate rack systems.Since its creation, the NVIDIA Hopper architecture has revolutionized AI and HPC, assisting researchers, developers, and businesses in resolving the most pressing problems in the world.
At Supercomputing 2024, the newest Hopper, the NVIDIA H200 NVL PCIe GPU, was introduced. For data center businesses looking for air-cooled, low-power enterprise rack designs with flexible configurations that can speed up AI or HPC applications of any size, the H200 NVL is the greatest choice.
Nearly 70% of corporate racks are 20kW or smaller and use air cooling, according to a recent survey. Thanks to PCIe GPUs, which provide granularity of node placement whether using one, two, four, or eight GPUs, data centers can now squeeze more computing power into less spaces. After that, companies can use their existing racks to select the number of GPUs that best suits their needs.
Companies can use H200 NVL to increase energy efficiency by consuming less electricity while also speeding up AI and HPC applications. Businesses may fine-tune LLMs in a few hours and attain up to 1.7x faster inference performance with the H200 NVL, which offers a 1.5x memory and 1.2x bandwidth gain over the NVIDIA H100 NVL. For HPC workloads, performance improvements of up to 1.3x over H100 NVL and 2.5x over the NVIDIA Ampere architecture generation are attained.
The raw power of the H200 NVL is increased by NVIDIA NVLink technology. The latest version of NVLink provides a GPU-to-GPU link seven times faster than fifth-generation PCIe to meet the needs of HPC, huge language model inference, and fine-tuning.
Businesses can use the NVIDIA H200 NVL to speed up applications from AI to HPC with the aid of powerful software tools. A five-year subscription includes NVIDIA AI Enterprise, a cloud-native software platform for developing and deploying production AI. NVIDIA AI Enterprise includes NVIDIA NIM microservices for the reliable and secure execution of high-performance AI model inference.
Businesses Leveraging the Potential of H200 NVL
With the H200 NVL, NVIDIA provides companies with a full-stack platform for creating and deploying AI and HPC applications.Customers are being significantly impacted by a variety of AI and HPC application cases across industries. Federal science groups, for instance, are utilizing visual AI agents and chatbots for customer service, pattern recognition for manufacturing, seismic imaging, medical imaging to enhance anomaly detection in healthcare, and trading algorithms for finance.
Dropbox uses NVIDIA accelerated computing for its services and infrastructure.
"Dropbox handles large amounts of content, requiring advanced AI and machine learning capabilities," said Ali Zafar, VP of Infrastructure at Dropbox. "It is looking into H200 NVL in order to keep improving its offerings and give customers more value."
The University of New Mexico has utilized NVIDIA accelerated computing for several academic and research endeavors.
"Being a public research university, the university is able to be at the forefront of scientific and technological advancements because of its dedication to AI." "A variety of applications will be able to accelerate as the transition to H200 NVL, including data science projects, bioinformatics and genomics research, physics and astronomy simulations, climate modeling, and more."
Available in All Ecosystems: H200 NVL
Hewlett Packard Enterprise, Lenovo, Supermicro, and Dell Technologies are expected to offer a range of configurations that support H200 NVL.Additionally, the H200 NVL will be provided by platforms from Aivres, ASRock Rack, ASUS, GIGABYTE, Ingrasys, Inventec, MSI, Pegatron, QCT, Wistron, and Wiwynn.
Certain systems are based on the NVIDIA MGX modular architecture, which enables computer manufacturers to quickly and affordably build a variety of data center infrastructure designs.
Platforms with H200 NVL will be made available by NVIDIA's global systems partners beginning in December. To enhance the products of leading international partners, NVIDIA is also developing an Enterprise Reference Architecture for H200 NVL systems.
The reference architecture will incorporate NVIDIA's design concepts and experience to help partners and customers develop and deploy high-performance AI infrastructure based on H200 NVL at scale. This provides full-stack hardware and software recommendations as well as detailed guidance on the optimal server, cluster, and network configurations. For optimum performance, networking is optimized by the NVIDIA Spectrum-X Ethernet platform.
NVIDIA technology will be on show on the exhibition floor during SC24, which is taking place at the Georgia World Congress Center through November 22.
0 Comments