The NVIDIA GH200 Grace Hopper Superchip is a breakthrough processor 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 CPU+GPU coherent memory model for accelerated workloads.
- Integrates NVIDIA Grace CPU and Hopper GPU architectures via NVLink-C2C
- Delivers 900 GB/s of coherent interface bandwidth, 7X faster than PCIe Gen5
- Features 72 power-efficient Arm-based Grace CPU cores
- Supports up to 624GB of combined CPU and GPU fast memory
- Utilizes HBM3 and HBM3e GPU memory for generative AI acceleration
- Provides up to 4 petaFLOPS of AI performance
- Enables CPU and GPU threads to access all system-allocated memory
- Accelerates data processing by up to 36X compared to traditional CPU-only systems
- Optimized for NVIDIA AI Enterprise, HPC SDK, and Omniverse software stacks
- Compatible with NVIDIA MGX modular reference designs for data centers
NVIDIA
NVIDIA - GH200 - Grace Hopper Superchip Multi-Node Cluster Module
For a quote, contact us at info@tropical.com
- SKU:
- GH200 Multi-Node Cluster Module
- Weight:
- 4.50 LBS
- Width:
- 4.40 (in)
- Height:
- 1.80 (in)
- Depth:
- 10.50 (in)
Related Products
A relatable product