GPU techniques

: novel approaches to deformation and fracture within videogames

Abstract

An investigation is introduced into methods of utilising current Graphics Processing Unit (GPU) features to migrate algorithms that are traditionally implemented on the Central Processing Unit (CPU).

The deformation, fracture and procedural edges sections of a destructible material simulation are implemented whilst developing novel solutions in order to improve performance and reduce content creation time.

The implementation takes the form of a simulation prototype and is specific to DirectX and HLSL, chosen because this target environment has enjoyed widespread use in the videogame demographic and is readily available on the PC platform although the implications of the work should be generalisable to systems running on other platforms or environments. The implementation also does not make use of General Purpose Graphics Processing Unit (GPGPU) Application Programming Interfaces (API’s) such as OpenCL [Khronos 2010a] and Direct- Compute [NVIDIA 2010b], or specialist configurations such as Nvidias PhysX [NVIDIA 2010c] or CUDA [NVIDIA 2011] that are targeted at hardware from a specific vendor as this research aims to encompass all compliant graphics hardware and allow the easiest possible adoption of any developed techniques, leveraging the large amount of knowledge within the videogame industry of the standard rendering pipeline and associated API’s.

The outcome demonstrates a destructible material simulation with a CPU and GPU implementation running on two different PC hardware configurations with both wood and rubber material types. The Frames Per Second (FPS) are recorded for comparison and the instability of the explicit Euler numerical integrator used is highlighted along with suggestions for improvement with future work.

A chapter is also provided covering some of the GPU best practices and tips for implementation of the GPU algorithms detailed, drawn from the experience gained whilst performing the required research to build the simulation prototype.

Date of Award	2010
Original language	English
Awarding Institution	Coventry University
Supervisor	Eike Anderson (Supervisor) & Christopher Peters (Supervisor)