CUDA Programming

Course details

This programme focusses on the CUDA (Compute Unified Device Architecture) parallel programming platform for NVidia GPUs. GPUs are the most common hardware acceleration being used today for High Performance Computing. GPUs outnumber any other acceleration mechanism in the Top500 list  the global ranking of the fastest supercomputers. With inexpensive GPU cards offering several TFlop/s of performance out-of-the-box, they are also the go-to method of choice to bring high performance to small, inexpensive systems. The course audience is anyone with experience in any number of programming languages who wants to utilise the power of GPU accelerators to develop high-quality and efficient parallel programs.

Please note that you must have a basic understanding of one of the following: C, C++, Java, Fortran, C#, or a similar high-level language.

Programme Aims and Objectives

This programme aims to provide the learner with a working knowledge of CUDA programming including memory management, both shared and unified virtual, and the ins-and-outs of parallel high-performance computing. The learner will be able to utilise any CUDA-enabled GPU to speed up operations like image processing, linear algebra and more.

Programme Content

The programme will be presented in a structured format by the trainer, and is principally and comprehensively geared towards preparing candidates for using CUDA in the workplace or in their own endeavours. Lab sessions and exercises will provide working examples of lecture content. The programme will expose the learner to the following core concepts and areas:

• GPU programming model and architecture
• GPU hardware and communication
• Parallel programming patterns and algorithms
• Optimising GPU programs
• GPU memory management
• Case studies (image processing, sorting, linear algebra, and more)
• Future directions of GPU programming