GPU Programming Part 1: Foundations

GPU-accelerated computing drives current scientific research. Writing fast numeric algorithms for GPUs offers high application performance by offloading compute-intensive portions of the code to a GPU.

The course will cover aspects of GPU architectures and programming. Focus is on the usage of the parallel programming language CUDA C++, which allows maximum control of NVIDIA GPU hardware. Examples of increasing complexity are used to demonstrate optimization and tuning of scientific applications.

The GPU Programming with CUDA course is held in two parts!

This current course is a basic course covering the foundations of GPU programming including an introduction to GPU/parallel computing, programming with CUDA, GPU libraries, tools for debugging and profiling, and performance optimizations.

In addition, an advanced course is available with modules providing more in-depth coverage of multi-GPU programming, modern CUDA concepts, CUDA Fortran, and portable programming models such as OpenACC and C++ parallel STL algorithms. The advanced modules will be taught from 3-7 June 2024, see the dedicated announcement for registration.

Contents of Part 1: Basics of GPU Programming with CUDA

Date: 8–10 April 2024 (this announcement)

A) Introduction to GPUs and GPU Computing
B) Programming Model CUDA
C) Tools for Debugging and Profiling
D) GPU Libraries (like cuBLAS, cuFFT)
E) Introduction to Multi-GPU Programming

Contents of Part 2: Advanced GPU Programming

Date: 3–7 June 2024 (see separate announcement)

A) Advanced Multi-GPU Programming with MPI
B) Advanced Multi-GPU Programming with NCCL and NVSHMEM
C) Advanced and Modern CUDA Concepts (Cooperative Groups, CUB Primitives, Modern C++ Programming)
D) CUDA Fortran
E) GPU Programming with Abstractions (OpenACC, Standard Language Programming (pSTL))

Attendees are invited to pick and choose the parts of the advanced course (A - E) they want to attend. The advanced modules are mostly freestanding. Participants either need to attend this basic course or prove equivalent knowledge of GPU programming in order to participate in the advanced course.

Programmers interested primarily in OpenACC may skip parts D and E of the basics course and still choose part E from Advanced GPU Programming. Participation in the full Basics of GPU Programming with CUDA course, however, is recommended.

Prerequisites:

Some knowledge about Linux, e.g. make, command line editor, Linux shell, experience in C/C++

Target audience:

Scientists who want to use GPU systems

Language:

This course is given in English.

Date:

8-10 April 2024, 09:00-17:00 each day

Venue:

tbd

Number of Participants:

maximum 26

Instructors:

Dr. Jan Meinke, Dr. Andreas Herten, Dr. Kaveh Haghighi-Mood, JSC;
Jiri Kraus, Markus Hrywniak, NVIDIA