Hardware-accelerated generation of 3D diffusion-limited aggregation structures

• Implementation of diffusion limited aggregation (DLA) on parallel hardware.
• Use of OpenCL running on multi-core CPU, GPU and FPGA.
• Performance evaluation of the accelerated DLA algorithm.

The diffusion and aggregation of particles in a medium can result in complex geometric forms with an artistic interpretation, yet these aggregates can represent many natural processes as well. Although the method is quite simple, it takes many particles to form an aggregation. If the process is simulated using a computer, it directly translates into lengthy computation times. In this paper, the acceleration of the diffusion-limited aggregation was investigated. The algorithm of aggregation was implemented on a serial single-core CPU, and that served as the base-case. With the aim of reducing run times, the algorithm was implemented on three accelerator architectures using OpenCL as the connecting software framework. Performance testing of the OpenCL implementation was done on a multi-core CPU, a GPU and an FPGA. Metrics such as run time, relative speedup and speedup-per-watt were used to compare the hardware-accelerated implementations. Even though using a GPU is not the most economical alternative energy-wise, its performance resulted in the highest speedup, while an FPGA or a multi-core CPU offered other viable options in accelerating the creation of diffusion-limited aggregation structures.