NVIDIA - GH200 - Grace Hopper Superchip for AI and HPC
NVIDIA - GH200 - Grace Hopper Superchip for AI and HPC

NVIDIA

NVIDIA - GH200 - Grace Hopper Superchip for AI and HPC

For a quote, contact us at info@tropical.com

SKU:
GH200 Grace Hopper Superchip
Weight:
3.50 LBS
Width:
7.68 (in)
Height:
7.28 (in)
Depth:
0.79 (in)
The NVIDIA GH200 Grace Hopper Superchip is a breakthrough accelerated CPU designed from the ground up for giant-scale AI and high-performance computing (HPC) applications. It integrates the NVIDIA Grace and Hopper architectures using NVIDIA NVLink-C2C to deliver a coherent memory model for seamless data processing. - Integrates NVIDIA Grace CPU and NVIDIA Hopper GPU architectures on a single module - NVIDIA NVLink-C2C interconnect provides 900 GB/s of coherent bidirectional bandwidth - Delivers up to 7X higher bandwidth than standard PCIe Gen5 interfaces - Equipped with high-bandwidth HBM3 or HBM3e GPU memory for generative AI workloads - Unified memory architecture allows CPU and GPU to share a single per-process page table - Features 72 power-efficient Arm Neoverse V2 CPU cores for high-throughput processing - Supports up to 624GB of total fast memory across the superchip - Provides up to 4 petaFLOPS of FP8 performance for AI training and inference - Designed for Retrieval-Augmented Generation (RAG) and Graph Neural Networks (GNN) - Compatible with NVIDIA MGX modular reference designs and NVIDIA AI Enterprise software

Related Products

A relatable product

!