Chapter 3

CUDA Hardware Overview

PC Architecture

Let’s start by looking at the typical Core 2 architecture we still find today in many PCs and how it impacts our usage of GPU accelerators (Figure 3.1).

image

Figure 3.1 Typical Core 2 series layout.

Notice that all GPU devices are connected to the processor via the PCI-E bus. In this case we’ve assumed a PCI-E 2.0 specification bus, which is currently the fastest bus available, giving a transfer rate of 5 GB/s. PCI-E 3.0 has become available at the time of this writing and should significantly improve the bandwidth available.

However, to get data from the processor, we need to go through the Northbridge ...

Get CUDA Programming now with the O’Reilly learning platform.

O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.