Reaching New Horizons, Hopper Speeds Up AI and HPC Applications for Commonplace Enterprise Servers.
NVIDIA H200 NVL
The NVIDIA H200 NVL is ideal for lower-power, air-cooled corporate rack systems because it enables 1.7x faster large language model inference and 1.3x faster high-performance computation.
Since its inception, the NVIDIA Hopper architecture has transformed AI and HPC, helping academics, developers, and companies solve the world’s biggest issues.
The NVIDIA H200 NVL PCIe GPU, the latest Hopper, was unveiled at Supercomputing 2024. The H200 NVL is the best option for data center enterprises seeking air-cooled, lower-power enterprise rack designs with adaptable configurations that can accelerate any size AI or HPC application.
A recent survey found that almost 70% of corporate racks utilize air cooling and are 20kW or less. Data centers can now fit more processing power into less places with to PCIe GPUs, which offer granularity of node placement whether utilizing one, two, four, or eight GPUs. After that, businesses may choose the amount of GPUs that best fits their requirements by using their current racks.
Businesses may utilize H200 NVL to speed up AI and HPC applications while simultaneously increasing energy efficiency by using less electricity. With a 1.5x memory and 1.2x bandwidth boost over the NVIDIA H100 NVL, businesses may fine-tune LLMs in a matter of hours and achieve up to 1.7x quicker inference performance with the H200 NVL. Performance gains of up to 1.3x over H100 NVL and 2.5x over the NVIDIA Ampere architecture generation are achieved for HPC applications.
NVIDIA NVLink technology enhances the H200 NVL’s raw power. To address the demands of HPC, massive language model inference, and fine-tuning, the most recent iteration of NVLink offers GPU-to-GPU connection seven times quicker than fifth-generation PCIe.
With the help of strong software tools, businesses may accelerate applications from AI to HPC with the NVIDIA H200 NVL. NVIDIA AI Enterprise, a cloud-native software platform for the creation and implementation of production AI, is included with a five-year subscription. For the safe and dependable implementation of high-performance AI model inference, NVIDIA AI Enterprise comes with NVIDIA NIM microservices.
Companies Tapping Into Power of H200 NVL
NVIDIA offers businesses a full-stack platform for developing and implementing their AI and HPC applications with H200 NVL.
Numerous AI and HPC use cases across industries are having a big impact on customers. For example, federal science organizations are using seismic imaging, medical imaging to improve anomaly detection in healthcare, pattern recognition for manufacturing, trading algorithms for finance, and visual AI agents and chatbots for customer service.
NVIDIA accelerated computing is being used by Dropbox for its infrastructure and services.
According to Ali Zafar, VP of Infrastructure at Dropbox, “Dropbox handles large amounts of content, requiring advanced AI and machine learning capabilities.” “In order to continuously enhance it offerings and provide to clients with greater value, its are investigating H200 NVL.”
NVIDIA accelerated computing has been used by the University of New Mexico for a number of scholarly and research purposes.
“As a public research university, the dedication to AI enables the university to be on the forefront of scientific and technological advancements.” “A number of applications, such as data science projects, bioinformatics and genomics research, physics and astronomy simulations, climate modeling, and more, will be able to accelerate as the transition to H200 NVL.”
H200 NVL Available Across Ecosystem
It is anticipated that Dell Technologies, Hewlett Packard Enterprise, Lenovo, and Supermicro will provide a variety of configurations that enable H200 NVL.
Furthermore, platforms from Aivres, ASRock Rack, ASUS, GIGABYTE, Ingrasys, Inventec, MSI, Pegatron, QCT, Wistron, and Wiwynn will provide the H200 NVL.
The NVIDIA MGX modular architecture, upon which certain systems are built, allows computer manufacturers to rapidly and economically construct a wide range of data center infrastructure designs.
Starting in December, NVIDIA’s international systems partners will provide platforms with H200 NVL. In addition, NVIDIA is creating an Enterprise Reference Architecture for H200 NVL systems to supplement the offerings of top international partners.
In order to enable partners and customers to build and implement high-performance AI infrastructure based on H200 NVL at scale, the reference architecture will integrate NVIDIA’s design principles and experience. This offers comprehensive advice on the best server, cluster, and network setups along with full-stack hardware and software recommendations. The NVIDIA Spectrum-X Ethernet platform optimizes networking for optimal performance.
During SC24, which is being held at the Georgia World Congress Center until November 22, NVIDIA technology will be on display on the exhibition floor.
