Models such as the Ising, Potts and heisenberg models in computational physics are still important tools for analysing phase transitionsand universal behaviours for new irregular and distrorted lattice networks. Data-parallelism can be exploited to speed up such simiulations as well as their analysis using general purpose graphical processing units and other accelerating devices. We report on the use of various GPU performance optimisation techniques for cluster-update algorithms such as those of Wolff and of Swendsen and Wang on such models. We present some data-parallel code fragments alongwith performance analysis and discussion of optimal GPU data memory models for this work. We employ NVIDIA's compute unified device architecture (CUDA) programming language but we also outline how recent language standards such as Open Compute Languaage (OpenCL) can also be used to achieve semi-portable high-performance simulations for phase transition simulations on irregular network structures.
Keywords: Ising model; cluster update; Wolff algorithm; CUDA, GPU; OpenCL.
Full Document Text: Not yet available.