News

NVIDIA GH200 AI server high-definition physical image

2023.12.19

6383573110107650608200030.jpg

GH200 Grace Hopper Superchip

6383573112324804736273434.jpg

72 Cores Grace CPU 


GH200 Grace Hopper Superchip Architecture 

6383573119649665259385055.png


Each Grace CPU has 72 ARM cores, equipped with 480GB LPDDR5X, with a memory bandwidth of up to 500GB/s.


NVIDIA H200 is the first GPU to offer HBM3e, a faster and larger memory that accelerates generative AI and large language models while enabling scientific computing for HPC workloads. With HBM3e, NVIDIA H200 delivers 141GB of memory at 4.8 TB per second.


The high-bandwidth NVLink-C2C link between the GH200 Grace Hopper Superchip NVIDIA Hopper GPU and Grace CPU enables fast communication between the CPU and GPU, which helps to improve performance.


The NVLink interface allows the topological connection of various GH200 components as building blocks for larger computing systems.


The NVIDIA Grace CPU supports up to 64 PCIe Gen 5 channels for IO connectivity. Each 8 PCIe Gen 5 x16 link supports up to 128 GB/s bi-directional bandwidth and can be divided into 2x8 for additional connectivity. It can support various PCIe slot form factors, including NVIDIA DPU, NVIDIA ConnectX SmartNIC, E1.S, and M.2 NVMe devices, modular BMC options, and more.


6383573122112235742874798.jpg

Supermicro NVIDIA MGX Server,with two NVIDIA GH200 Grace Hopper Superchip

Note: The above content is collated from the Internet, and the copyright belongs to the original author. If there is any infringement, please contact us for deletion.


Prev: The challenge of 800G switch to PCB interconnection

Next: Why rigid-flex PCB is more suitable for portable electronic devices?

Back
Subsidiaries:SGC Germany  |  Sunshine PCB GmbH   |  SGC USA
©Sunshine Global Circuits Co., Ltd.2018 All rights reserved. 粤ICP备05084072号   Design byszweb.cn