MeluXina: Introduction to GPU Programming Using CUDA

This course is aimed at both current and prospective users of large hybrid CPU/GPU clusters and supercomputers. If you are involved in developing or parallelizing scientific computing applications using GPUs, we encourage you to participate!

Participants in this course will learn GPU programming using the CUDA programming model, which includes synchronization, memory allocation, and device and host calls. You will gain an understanding of GPU architecture and how parallel thread blocks are utilized to parallelize computational tasks. As GPUs act as accelerators, a solid grasp of memory management between the GPU and CPU will also be covered extensively. Additionally, participants will learn how to use the CUDA programming model to accelerate linear algebra routines and iterative solvers on the GPU. The course will present theoretical concepts first, followed by hands-on implementation of the CUDA programming model with guidance from mentors.

After completing the course, participants will be able to:

      1.  Understand GPU architecture, including the differences between GPU and CPU:

                  - Streaming architecture 

                  - Thread blocks

      3.  Efficiently manage memory, covering: 

                  - Host to Device transfer

      4.  Apply CUDA programming knowledge to accelerate examples from science and engineering, including: 

                  - Iterative solvers

                  - Matrix multiplication, vector addition, etc.

Prerequisites 

Preference will be given to users with a solid background in C/C++. No prior GPU programming knowledge is required, but familiarity with basic parallel programming concepts is advantageous.

Participants will be granted access to the MeluXina supercomputer during the session. For more information about MeluXina, please refer to the MeluXina overview and the MeluXina Getting Started Guide.