Parallel Numerical Methods

2-0-4 units · Undergraduate

Focuses on concepts and methods for parallelizing numerical computations, with concrete examples drawn from a variety of disciplines, including physical models and machine learning. Parallelization techniques include shared-memory threading, distributed-memory message passing, and dynamic scheduling of irregular problems described by computational graphs, from laptop CPUs to GPU-enabled supercomputer clusters. Software examples employ the Julia language and libraries such as MPI and Dagger.jl. Includes a final project. Second half-term subject building on basic ideas introduced in 18.C21 .