NVIDIA announces that the NVIDIA H100 Tensor Core GPU is in full production, and global technology partners plan to launch the first wave of products and services based on the revolutionary NVIDIA Hopper™ architecture in October.
Unveiled in April, the H100 is built with 80 billion transistors and features a number of technological improvements. These include the powerful new Transformer Engine and NVIDIA NVLink® interconnect to accelerate the largest artificial intelligence (AI) models, such as advanced recommender systems and large language models, and drive innovation in fields such as conversational artificial intelligence and drug discovery.
“Hopper is the new engine of AI factories that process and refine mountains of data to train models with trillions of parameters used to drive advances in language-based AI, robotics, healthcare and life sciences,” said Jensen Huang, founder and CEO of NVIDIA. “Hopper’s Transformer Engine increases performance by an order of magnitude, putting large-scale AI and HPC within reach of enterprises and researchers.”
In addition to the Hopper architecture and Transformer Engine, several other key innovations power the H100 GPU for the next big leap in NVIDIA’s accelerated data center computing platform, such as a second-generation multi-instance GPU, confidential computing, fourth-generation NVIDIA NVLink generation, and DPX instructions .
A five-year license for the NVIDIA AI Enterprise software suite is now included with the H100 for mainframe servers. This guides the development and implementation of AI workflows and provides organizations with access to the AI frameworks and tools they need to build AI chatbots, recommendation engines, visual AI and more.
“The NVIDIA Hopper architecture is a major advancement for accelerated computing and these new features bring even more benefits to diverse enterprise and data center workloads,” added Marcio Aguiar, director of NVIDIA Enterprise Latin America.
Hopper global launch
The H100 enables businesses to dramatically reduce AI deployment costs by delivering the same AI performance with 3.5x higher energy efficiency and 3x lower total cost of ownership, while using 5x fewer server nodes compared to the previous generation.
For customers looking to try the new technology immediately, NVIDIA announced that the H100 on Dell PowerEdge servers is now available on the NVIDIA LaunchPad, which offers free hands-on labs, giving businesses access to the latest NVIDIA hardware and AI software.
Customers can also start ordering NVIDIA DGX™ H100 systems, which include eight H100 GPUs and deliver 32 petaflops of FP8 accurate performance. NVIDIA Base Command™ and NVIDIA AI Enterprise software power all DGX systems, enabling single-node deployments on NVIDIA DGX SuperPOD™ that support advanced AI development of large language models and other massive workloads.
H100 systems from the world’s leading PC manufacturers are expected to launch in the coming weeks, with more than 50 server models on the market by the end of the year and dozens more in the first half of 2023. System manufacturing partners include Atos, Cisco, Dell Technologies, Fujitsu, GIGABYTE , Hewlett Packard Enterprise, Lenovo and Supermicro.
In addition, some of the world’s leading higher education and research institutions will use the H100 to power their state-of-the-art supercomputers. Among them, Barcelona Supercomputing Center, Los Alamos National Laboratory, Swiss National Supercomputing Center (CSCS), Texas Advanced Computing Center and University of Tsukuba.
H100 comes to the cloud
Amazon Web Services, Google Cloud, Microsoft Azure and Oracle Cloud infrastructure will be among the first to deploy H100-based instances in the cloud starting next year.
“We look forward to enabling the next generation of AI models on the latest Microsoft Azure H100 GPUs,” said Nidhi Chappell, general manager of Azure AI Infrastructure. “With advances in the Hopper architecture along with our investments in Azure AI supercomputers, we will be able to help accelerate the development of AI around the world.”
“We offer our customers the latest NVIDIA H100 GPUs to help them accelerate their most complex machine learning and HPC workloads,” said Karan Batta, vice president of product management, Oracle Cloud Infrastructure. “Additionally, using NVIDIA’s next-generation H100 GPU allows us to support our demanding internal workloads and help our mutual customers with advances in healthcare, autonomous vehicles, robotics and IoT.”
NVIDIA software support
The advanced technology of the Transformer Engine H100 enables companies to rapidly develop large language models with a higher level of accuracy. As these models continue to grow in scale, so does complexity, sometimes requiring months of training.
To address this, the world’s leading deep learning language models and frameworks have been optimized on the H100, including NVIDIA NeMo Megatron, Microsoft DeepSpeed, Google JAX, PyTorch, TensorFlow and XLA. These frameworks combined with the Hopper architecture will significantly accelerate AI performance to help train large language models in days or hours.
To learn more about the NVIDIA Hopper and H100, watch Huang’s presentation at GTC.
Since its founding in 1993, NVIDIA (NASDAQ: NVDA ) has been a pioneer in accelerated computing. The company’s invention of the GPU in 1999 fueled the growth of the PC gaming market, redefined computer graphics and ushered in the era of modern artificial intelligence. NVIDIA is now an end-to-end computing company with data center-level offerings that are reshaping the industry. More information at: https://www.nvidia.com/pt-br/.
Certain statements in this press release including, but not limited to, statements about: the benefits, impact, specifications, performance, features and availability of our products and technologies, including NVIDIA H100 Tensor Core GPUs, NVIDIA Hopper Architecture, NVIDIA AI Enterprise software package , NVIDIA LaunchPad, NVIDIA DGX H100 systems, NVIDIA Base Command, NVIDIA DGX SuperPOD and NVIDIA Certified Systems; a number of the world’s leading computer manufacturers, cloud service providers, higher education and research institutions, as well as major language and deep learning model frameworks that embrace H100 GPUs; software support for NVIDIA H100; large language models continue to grow; and the performance of large language models and deep learning frameworks combined with the NVIDIA Horizontal Architecture are forward-looking statements that are subject to risks and uncertainties that could differ materially from expectations. Among the important factors that could cause actual results to differ materially are global economic conditions; our reliance on third parties to manufacture, assemble, package and test our products; impact of technological development and competition; development of new products and technologies or improvement of existing ones; market acceptance of our products or those of our partners; software, design or manufacturing defects; changes in consumer preferences or requirements; changes in interfaces and industry standards; unexpected loss of performance of our products or technologies when integrated with systems; and other factors regularly disclosed in NVIDIA’s most recent filings with the Securities and Exchange Commission (SEC), including, but not limited to, the Annual Report on Form 10-K and Quarterly Reports on Form 10-Q. Copies of the reports filed with the SEC are posted on the company’s website and are available free of charge from NVIDIA. These forward-looking statements are not guarantees of future performance and are based on facts as of the date hereof, and, except as required by law, NVIDIA disclaims any obligation to update such statements to reflect future circumstances or events.
© 2022 NVIDIA Corporation. All rights reserved. NVIDIA, the NVIDIA logo, DGX, NVIDIA Base Command, NVIDIA-Certified Systems, NVIDIA DGX SuperPOD, NVIDIA Hopper and NVLink are trademarks and/or registered trademarks of NVIDIA Corporation in the US and other countries. Other company and product names are trademarks of the respective companies with which they are associated. Features, pricing, availability and specifications are subject to change without notice.