Gpu Architecture & Programming

This course introduces Graphics Processing Unit (GPU) architectural building blocks such as global, constant, texture, scratchpad, and cache memory. Lectures center around the GPU massive parallelism concept and techniques in building optimum-performance programs in GPU platforms by comparing CPU and GPU platforms. Although the course primarily utilizes the widely used Compute-Unified Device Architecture (CUDA) GPU programming language, it also introduces the Open-CL language to compare and contrast syntactic and performance differences. Lectures teach methodologies in using CUDA to implement parallel sorting, reduction, numeric iterations, and fundamental graphics operations, such as ray tracing in a manner that progressively increases in sophistication and performance. The course incorporates class-wide literature reviews and discussion sessions to study key publications that introduce and detail GPU architecture and programming. Offered by Electrical & Comp. Engineering. May not be repeated for credit.