<aside> ❗ Headings and text with “▶” can be toggled to expand contained info. You can also see the Table of Contents as a collapsible sidebar on the right →

</aside>

Table of Contents

Introduction

This module introduces the hardware architecture of multi-GPUs and provides an overview of several methods for increasing concurrency on GPUs.

Objective

• Understand multi-GPU architecture
• Understand the challenges of multi-GPU architectures
• Understand how to increase concurrency on multi-GPUs

Readings

Required

Optional

Notes / Things that stand out

Multi-GPU Hardware

Multi-GPUs

• To accommodate high demand for computing power, GPUs have expanded into multi-GPU systems.
• First, GPUs have evolved by increasing the number of cores.
  • However, this monolithic GPU approach would face several challenges, including cost and reliability.
• To overcome said challenges, the next step is to scale using silicon interposers, or multi-sockets.
• After that, we can even scale up by connecting these GPU boards with multiple PCI-E connections.

Multi-GPU Connections