Harnessing trillion-parameter AI models, NVIDIA launches Blackwell-based DGX SuperPOD supercomputer

According to news from this site on March 19, at the GTC Developer Conference held today, NVIDIA launched the DGX SuperPOD for processing trillion-parameter AI models, providing stability for ultra-large-scale generative AI training and inference workloads. performance output.

This site learned from reports that the new DGX SuperPOD adopts a new high-efficiency water-cooled rack-mounted architecture and is equipped with the NVIDIA DGX GB200 system, providing 11.5 exaflops at FP4 precision. Artificial intelligence supercomputing power and 240 terabytes of fast memory, customers can further expand by adding racks.

Each DGX GB200 system has 36 NVIDIA GB200 accelerator cards, including 36 NVIDIA Grace CPUs and 72 NVIDIA Blackwell GPUs, connected into a supercomputer through fifth-generation NVIDIA NVLink.

The GB200 accelerator card delivers up to 30x better performance on large language model inference workloads compared to the NVIDIA H100 Tensor Core GPU.

DGX SuperPOD contains 8 or more DGX GB200 systems, connected through NVIDIA Quantum InfiniBand, and can be expanded to tens of thousands of GB200 accelerator cards.

NVIDIA’s current default configuration for customers includes 576 Blackwell GPUs connected via NVLink to eight DGX GB200 systems.

Jensen Huang, founder and CEO of NVIDIA, said:

NVIDIA DGX artificial intelligence supercomputer is the factory of the artificial intelligence industry revolution. The new DGX SuperPOD combines the latest advances in NVIDIA accelerated computing, networking and software to enable every company, industry and country to refine and generate their own artificial intelligence.

The above is the detailed content of Harnessing trillion-parameter AI models, NVIDIA launches Blackwell-based DGX SuperPOD supercomputer. For more information, please follow other related articles on the PHP Chinese website!