Advanced Gpu Programming And Deep Learning

This course expands on the GPU architecture and programming concepts introduced in ECE 555 by detailing advanced architectural components, such as Tensor Processing Units (TPUs), and their role in accelerating Deep Learning (DL) applications. Lectures study example DL applications, such as object recognition, and how the GPU instruction sets, e.g., Parallel Thread Execution (PTX), are improved to accelerate DL by utilizing TPUs, load/store controllers, fetch, decode, and execute cycles, cache utilization, and emerging data types such as half-precision floating-point. Literature review sessions and lectures facilitate a rigorous study of the data flow in and out of the GPU to determine optimal ways to provide hardware acceleration for data-intensive deep learning using GPUs. Concepts such as Multi-GPU execution and virtual addressing are also introduced for further performance improvements and code modularization. Offered by Electrical & Comp. Engineering. May not be repeated for credit.