ECE 532GPU Programming

Southern Illinois University Carbondale · 3 credits

Catalog

This course provides an advanced study of GPU architectures and high-performance parallel programming using CUDA C/C++. Students will explore many-core design principles in depth, focusing on low-level architectural features such as warp schedulers, SIMT execution, shared memory banks, and memory access patterns. Emphasis is placed on performance-critical optimizations, including instruction-level parallelism, occupancy tuning, register pressure analysis, and efficient tensor core usage for AI and scientific computing. The course also covers profiling and fine-grained performance analysis using tools such as Nsight Compute and CUDA Profiler. Students will work with advanced CUDA libraries such as cuDNN, cuBLAS, and CUTLASS to implement and optimize compute-intensive kernels. Design trade-offs in cache hierarchy, memory bandwidth, and interconnects will be examined in the context of throughput, scalability, and energy efficiency. Students that have completed ECE 432 are not eligible to enroll in ECE 532 as the courses cover similar topics to an extent. Project-based fee: $20 to help defray cost of equipment.

Recent sections

Spring 2027 · 2 sections

SectionDays & timeLocationInstructor
941Class #24694Time not publishedJan 11 – May 9Off-Campus 62901Iraklis Anagnostopoulos
001Class #25152Mo We Fr · 4:00 – 4:50 PMJan 11 – May 7Engineering, E Wing 0136Iraklis Anagnostopoulos