NVIDIA Hopper H100 GPU: Scaling Performance

被引：30

作者：

Choquette, Jack ^{[1
]}

机构：

[1] NVIDIA, Santa Clara, CA 95051 USA

来源：

IEEE MICRO | 2023年 / 43卷 / 03期

关键词：

Graphics processing units; Instruction sets; Tensors; Memory management; Artificial intelligence; Transforms; Bandwidth;

D O I：

10.1109/MM.2023.3256796

中图分类号：

TP3 [计算技术、计算机技术];

学科分类号：

0812 ;

摘要：

The H100 Tensor Core GPU is NVIDIA's latest flagship GPU. It has been designed to provide industry leading performance for high-performance computing, artificial intelligence, and data analytics datacenter workloads. Notable new features include a fourth-generation Tensor Core, new Tensor Memory Accelerator unit, a new CUDA cluster capability, and HBM3 dynamic random-access memory.

引用

页码：9 / 17

页数：9