CUDA Advanced Libraries

Johns Hopkins University via Coursera

Coursera based on 0 ratings

Likes

Overview

The course “CUDA Advanced Libraries” delves into the powerful ecosystem of NVIDIA CUDA libraries designed to accelerate high-performance computing applications. Building on foundational knowledge of CUDA programming, this course focuses on advanced libraries that simplify the development of GPU-accelerated software across domains such as scientific computing, deep learning, computer vision, and data analytics. Participants will explore key CUDA libraries including cu BLAS for optimized linear algebra operations, cu FFT for fast Fourier transforms, cu DNN for deep neural networks, and...

Syllabus

Course Overview

The purpose of this module is for students to understand how the course will be run, topics, how they will be assessed, and expectations.

cuFFT

cuFFT provides the ability to perform fast Fourier transforms (FFTs) on large datasets. Students will learn of common use cases such as fast multiplication of large polynomials, signal processing, and matrix operations. They will use this library to develop software that process audio or video signals.

CUDA Linear Algebra

The CUDA Toolkit includes a number of linear algebra libraries, such as cuBLAS, NVBLAS, cuSPARSE, and cuSOLVER. Students will learn the different capabilities and limitations of many of them and apply that knowledge to compute matrix dot products, determinant, and finding solutions to complex linear systems.

The CUDA Thrust Library

Most developers utilize data structures beyond the primitives and pointers that make up the core of CUDA programmers, which makes pure CUDA development difficult. Students will learn about the Thrust library that adds the vector data structure and associated algorithms that allow for simplification of their code. Students will create software that transform, reduction, and sort large datasets.

CUDA Machine Learning

Data Scientists, Machine Learning, and Artificial intelligence experts are using neural networks to solve problems such as human language translation, image classification, and object detection/avoidance. Using the cuDNN and cuTensor, students will be able to develop a variety of neural networks and similar structures. At the completion of this module students will be asked to develop a course-wide project that brings together their knowledge from all previous courses and lessons to develop a capstone software project.