Fast Rendering of Dynamic Procedural Content on a GPU
Proposer: Patrick Marais
Brief Description: Procedural content can be used to reduce the storage requirements of game assets such as trees, debris, terrain, buildings and so on. Currently, most procedural game assets are generated prior to game start-up, and a limited number of instances are utilised in-game. One can think of this as ‘static’ procedural content. We are interested in the extent to which procedural elements can be generated on the fly, and the computational and rendering trade-offs this involves. To investigate these issues, we propose selecting a particular procedural type - such as terrain – and implementing and evaluating a real-time procedural rendering system for this type. A full performance assessment and analysis will provide useful information with regards to the extent to which this approach to game world generation is feasible. Modern GPUs provide a rich features-set which can help to realise these goal, and the rendering engine will need make full use of new generation functionality.
Computer Science Content: Computer Graphics - procedural generation, GPU programming, algorithms
Specific Learning Outcomes: GPU algorithm design; design of efficient rendering data structures and algorithms
Skills Required by Team as a Whole:
Theory: Computer graphics; data structures and algorithms; render-engine design; performance analysis
Implementation: hard - willingness to learn GPU coding and the techniques used to facilitate efficient display of large scenes in real-time.
Facilities needed: a DX10/11 capable card and a high-end PC
Supervision: James Gain may serve as co-supervisor if he is available. Students will be required to attend regular supervisory meetings and to meet agreed milestones.
Number of Students: 3