From NVIDIA:

The NVIDIA GTC conference will feature accelerated systems with powerful processors named after Grace Hopper. Over 500 NVIDIA GH200 Grace Hopper Superchips will be showcased in various configurations at the event. The systems support various AI, HPC, and NVIDIA Omniverse applications and are part of the NVIDIA MGX modular server blueprint.

System makers will display servers with GH200 Superchips at GTC, including ASRock RACK’s MECAI, ASUS’s ESC NM2N-E1, and Foxconn’s suite of MGX systems among others. These systems offer accelerated AI and 5G services, support multiple GH200 processors, and accommodate various GPU configurations and data processing units.

The GH200 Superchips combine a Grace CPU and NVIDIA H100 GPU, sharing memory over an NVLink-C2C interconnect for high performance. These processors are well-equipped for demanding tasks like running large language models and generative AI. The GH200 delivers increased efficiency and up to 4x more performance compared to traditional CPU-GPU setups.

GH200 systems have made an impact on the MLPerf benchmarks, running data center inference tests with leading performance. These systems build on Grace Hopper’s legacy of computing revolution, offering increased performance and efficiency for tasks like recommendation systems. Get a closer look at these systems at the NVIDIA GTC conference.



Read more at NVIDIA: Rack ‘n’ Roll: NVIDIA Grace Hopper Systems Gather at GTC