<p>Lecture, four hours; discussion, two hours; outside study, six hours. Requisite: course M151B. Exploration of design principles, optimization strategies, and emerging trends in graphics processing unit (GPU) microarchitecture. Emphasis on their critical role in artificial intelligence, machine learning, and graphics rendering. Through a blend of theoretical analysis, hands-on projects, and cutting-edge research discussions, study dissects the architectural innovations driving modern GPUs, such as Nvidia’s tensor cores and ray tracing. Leverages the open-source Vortex GPU platform for practical experimentation, enabling students to implement and evaluate microarchitectural concepts in Verilog. Topics include compute/graphics pipelines, memory hierarchies, parallelism, compiler optimizations, and hardware/software codesign. Students propose and execute a research project, supported by curated readings from patents, seminal papers, and industry white papers. Letter grading.</p>

Computer Science

Henry Samueli School of Engineering and Applied Science

Container 1

COM SCI 254A - Advanced Graphics Processing Unit Microarchitecture for Artificial Intelligence and Graphics Rendering

Advanced Graphics Processing Unit Microarchitecture for Artificial Intelligence and Graphics Rendering

Analyze the trade-offs in GPU compute vs. graphics pipelines and their impact on AI workloads.

Evaluate memory virtualization techniques and compiler optimizations for parallelism.

Develop performance/power models for modern GPU architectures

Design and simulate GPU components (e.g., warp schedulers, tensor cores) using Verilog and Vortex GPU.

Understand the unique features of GPU microarchitectures and their critical role in scaling GPU performance and efficiency.

General Catalog

Advanced Graphics Processing Unit Microarchitecture for Artificial Intelligence and Graphics Rendering

COM SCI 254A

4 units