NVIDIA has announced the launch of its fourth-generation NVIDIA DGX system, an AI platform built with new NVIDIA H100 Tensor Core GPUs. The scale required to handle the vast computing requirements of big language models, recommender systems, healthcare research, and climate science is provided by DGX H100 systems.
Each DGX H100 has eight NVIDIA H100 GPUs connected as one via NVIDIA NVLink, delivering 32 petaflops of AI performance at new FP8 precision, which according to NVIDIA is 6x faster than the previous generation.
The next-generation NVIDIA DGX POD and NVIDIA DGX SuperPOD AI infrastructure solutions are built on DGX H100 systems. The new NVIDIA NVLink Switch System in the DGX SuperPOD architecture can link up to 32 nodes with a total of 256 H100 GPUs.
The next-generation DGX SuperPOD stretches the boundaries of AI with the capacity to handle enormous LLM workloads with billions of parameters, providing 1 exaflops of FP8 AI performance, 6x more than its predecessor, stated NVIDIA.
“AI has fundamentally changed what software can do and how it is produced. Companies revolutionizing their industries with AI realize the importance of their AI infrastructure,” said Jensen Huang, Founder and CEO of NVIDIA. “Our new DGX H100 systems will power enterprise AI factories to refine data into our most valuable resource – intelligence.”
Announcing NVIDIA Eos – AI Supercomputer
NVIDIA will be the first to develop a DGX SuperPOD with the ground-breaking new AI architecture, which will fuel NVIDIA researchers’ efforts in climate science, digital biology, and AI’s future.
Its ‘Eos’ supercomputer, which will be operational later this year, is intended to be the world’s fastest AI system, with 576 DGX H100 computers and 4,608 DGX H100 GPUs.
NVIDIA Eos is expected to produce 18.4 exaflops of AI computing capability, which is 4x faster than Japan’s Fugaku supercomputer, which is now the fastest machine in the world. Eos is intended to deliver 275 petaflops of performance for typical scientific computing. Eos will serve as a template for NVIDIA’s advanced AI infrastructure, as well as its OEM and cloud initiatives.
DGX H100 Systems: Enterprise AI Scaling
As businesses develop from pilot studies to large-scale deployments, DGX H100 systems would simply scale to meet the needs of AI.
Each DGX H100 system incorporates two NVIDIA BlueField-3 DPUs to offload, accelerate, and isolate sophisticated networking, storage, and security services, in addition to eight H100 GPUs with a total of 640 billion transistors.
Eight NVIDIA ConnectX-7 Quantum-2 InfiniBand networking adapters deliver 400 gigabits per second throughput for computation and storage, which is twice as fast as the previous generation system. And, when paired with NVSwitch, a fourth-generation NVLink enables 900 gigabytes per second communication between each GPU in each DGX H100 system, 1.5 times faster than the previous generation.
DGX H100 systems use dual x86 CPUs and can be combined with NVIDIA networking and storage from NVIDIA partners to make flexible DGX PODs for AI computing at any size.
DGX SuperPOD provides a scalable enterprise AI center of excellence with DGX H100 systems. The DGX H100 nodes and H100 GPUs in a DGX SuperPOD are connected by an NVLink Switch System and NVIDIA Quantum-2 InfiniBand providing a total of 70 terabytes/sec of bandwidth – 11x higher than the previous generation. Storage from NVIDIA partners will be tested and certified to meet the demands of DGX SuperPOD AI computing.
Multiple DGX SuperPOD units can be combined to provide the AI performance needed to develop massive models in industries such as automotive, healthcare, manufacturing, communications, retail and more.
Starting in the third quarter, NVIDIA’s global partners will provide DGX H100 systems, DGX PODs, and DGX SuperPODs.
Clients may also install DGX systems at Cyxtera, Digital Realty, and Equinix IBX colocation data centers, which are also NVIDIA DGX-Ready Data Center partners.