Lecture 07: Towards Parallel Computing

Content locked

The catalog is public. Reading this content requires approved access.

Announcements

Code of the day: remove one set bit at a time

Why sequential performance hits three walls

Memory: latency and bandwidth are different

ILP: complicated hardware is not unlimited parallel work

Power: useful work per watt

From a few strong cores to many simpler workers

Multicore CPU: Haswell as a historical example

GPU: Fermi as a historical example

Flynn’s taxonomy

Amdahl’s law: how much can parallelization help?

First C++ thread example

Shared memory creates correctness problems

Producer and consumer

Two-store, two-load thought experiment

Where parallelism can be expressed