In this course, you will learn how high-performance parallel architectures are designed and how they are programmed. Especially, the (micro-)architectural aspects of GPGPU (General-Purpose computing on Graphics Processing Units) and CUDA programming will be extensively covered in the first half of the course. The other half will cover various topics in parallel systems including domain-specific accelerators, memory system, interconnection networks, large-scale systems and emerging paradigms. Students will also have hands-on experiences on CUDA programming through assignments.

CSED311 (Computer Architecture)

(subject to change)

Mid-term exam (40%)
Final exam (40%)
Programming assignments (20%)

(This is a tentative schedule and subject to change)
Week 1: Course introduction, GPU microarchitecture
Week 2: CUDA basics, Memory and locality
Week 3: Threads and atomics, performance considerations
Week 4: Parallel patterns – convolution, tiled convolution
Week 5: Parallel patterns – reduction
Week 6: Parallel patterns – parallel scan
Week 7: Parallel patterns – histogram, parallel sparse methods
Week 8: Midterm exam
Week 9: Advanced CUDA
Week 10: Deep learning primer, deep learning with GPU
Week 11: Efficient deep learning
Week 12: Deep learning with domain-specific accelerators
Week 13: Interconnection networks
Week 14: Advanced topics on parallel architecture
Week 15: Advanced topics on parallel programming
Week 16: Final exam

Lectures and assignments
For the first two weeks of the semester, lectures will be given through Zoom (meeting information will be made available on PLMS) due to the university policy regarding COVID-19.
If you haven't enrolled in the course but plan to do so, please send an audit request through PLMS.