SYLLABUS PAGE, 2009/10

06-N0097
Programming Massively Parallel Architectures

Level 4/M

Dr D R Ghica

10 credits in Sem1

The School of Computer Science Module Description is a strict subset of this Syllabus Page. (The University module description has not yet been checked against the School's.)

Changes and Updates

New module for 2009/10; awaiting module code.

Relevant Links

Relevant link.

Outline

This module covers the basics of programming massively parallel processors as commonly found in graphics processing units (GPUs). The module is focussed on the architecture of such devices and their use in speeding-up common (non-graphical) computational tasks. Recent developments in GPU architectures made this task substantially easier and newly available programming tools allow for high-level straightforward programming of these systems. The module will provide a largely practical introduction to the topic. Lectures will address the basic architectural principles of modern GPUs andrelevant algorithms and programming techniques. The exercises will be practical programming based on the CUDA framework for Nvidia GPUs.

Aims

The aims of this module are to:

Understand GPU architecture.
Understand what applications areas are most suitable to parallel programming.
Practical understanding of parallel algorithms.
Practical understanding of parallel programming techniques.
Concrete skills in programming Nvidia GPUs using the CUDA framework.
Skills in using the software tools provided by the CUDA framework.

Learning Outcomes

On successful completion of this module, the student should be able to:		Assessed by:
1	Describe and explain modern GPU architecture.	Examination
2	Describe and explain applications of parallel programming.	Examination
3	Describe and explain the CUDA programming model.	Examination
4	Design simple parallel algorithms.	Examination and continuous assessment
5	Implement more advanced parallel algorithms using CUDA.	Continuous assessment
6	Use CUDA tools to debug and profile programs.	Continuous assessment

Restrictions, Prerequisites and Corequisites

Restrictions:

For 2009/10, the module can only accommodate a certain maximum number of students, depending on how much equipment can be purchased. If too many students opt for this module, a test will be conducted to limit the numbers of students.

Prerequisites:

None

Co-requisites:

None, but systems level programming skills will be helpful.

Teaching

Teaching methods:

2 hrs/week lectures; 2 hrs/week practical sessions

Contact hours:

Assessment

Normal (sessional): Normal (sessional): 1.5 hr examination (80%), continuous assessment (20%)

Resit (supplementary) assessment (where allowed): By examination only (100%)

Recommended Books

Title	Author(s)	Publisher, Date	Comments
Multigrid Methods on GPUs	Peter Thoman	VDM Verlag,
GPU Gems 3: Programming Techniques for High-Performance Graphics and General-Purpose Computation	Hubert Nguyen	Addison Wesley,

Detailed Syllabus

The detailed syllabus is subject to change.

There will be a set of weekly practicals.

Contrasting CPU and GPU architectures
The CUDA programming model
A first example: matrix multiplication
The CUDA memory model
GPU as part of the PC architecture
Detailed threading model
Detailed memory model
Control flow on the GPU
Floating-point aspects
Parallel programming concepts
Parallel algorithms
Reduction
Advanced features

06-N0097Programming Massively Parallel Architectures