NVIDIA A100 Tensor Core GPU

A100多实例GPU( NVIDIA Multi-Instance GPU(MIG) )

A100 for NVLink vs. for PCIe

NVIDIA A100 Tensor Core GPU有两种接口形式:

• NVIDIA SXM Socket(接口) 提供了 2.039 GB/s GPU内存带宽

• PCIe 版本的GPU内存带宽为 1.555 GB/s

从技术规格来看，两者主要的差别是GPU Memory (Bandwidth): SXM Socket接口更胜一筹，而且SXM配置的GPU Memory能够达到80GB(也有40GB 规格)，而PCIe接口只有40GB规格一种。

参考

• NVIDIA A100 Tensor Core GPU: Unprecedented acceleration at every scale 官方网站A00产品Portal

• NVIDIA A100 TENSOR CORE GPU: UNPRECEDENTED SCALE AT EVERY SCALE 官方A100产品介绍pdf

• NVIDIA A100 SXM4 80 GB

• NVIDIA A100 SXM GPU 40GB and 80GB - System Overview

• NVIDIA A100 Concepts, Specs, Features 搜集A100规格非常详细