Term of the Moment
KyotoCooling
Definition: NVIDIA Grace Hopper Superchip
Named in honor of Rear Admiral Grace Hopper, the NVIDIA superchip was designed for AI and high-performance computing. Introduced in 2023, the superchip contains NVIDIA's ARM-based Grace datacenter CPU with 72 cores and the Hopper H100 GPU that connect to each other via NVLink at 900 gigabytes per second. See NVIDIA DGX.
Grace Hopper Superchip
Commodore Grace Hopper
A Note from the Author
As a program director for computer seminars at the American Management Association in New York in the 1970s, I had the privilege of knowing Grace Hopper. On several occasions, I handled the logistics for the speeches she gave and escorted her to the venues. She had a very dry sense of humor, and wherever she went in her official naval uniform, she raised eyebrows.
The Grace CPU connects to 480GB of low power (LPDDR RAM and to the outside world via four 16-lane PCIe 5 channels. The Hopper GPU uses 18 NVLinks to a network of GPUs and connects to high bandwidth memory (96GB HBM and 144GB HBM3e). See LPDDR SDRAM, PCI Express and high bandwidth memory.
One of the most famous people in the early history of computing, Commodore Grace Hopper became a rear admiral when the rank of commodore was renamed. See Mark I and FLOW-MATIC. (Image courtesy of U.S. Navy.)
As a program director for computer seminars at the American Management Association in New York in the 1970s, I had the privilege of knowing Grace Hopper. On several occasions, I handled the logistics for the speeches she gave and escorted her to the venues. She had a very dry sense of humor, and wherever she went in her official naval uniform, she raised eyebrows.