Exploiting graphical processing units for data‐parallel scientific applications

Abstract

SUMMARYGraphical processing units (GPUs) have recently attracted attention for scientific applications such as particle simulations. This is partially driven by low commodity pricing of GPUs but also by recent toolkit and library developments that make them more accessible to scientific programmers. We discuss the application of GPU programming to two significantly different paradigms-regular mesh field equations with unusual boundary conditions and graph analysis algorithms. The differing optimization techniques required for these two paradigms cover many of the challenges faced when developing GPU applications. We discuss the relevance of these application paradigms to simulation engines and games. GPUs were aimed primarily at the accelerated graphics market but since this is often closely coupled to advanced game products it is interesting to speculate about the future of fully integrated accelerator hardware for both visualization and simulation combined. As well as reporting the speed-up performance on selected simulation paradigms, we discuss suitable data-parallel algorithms and present code examples for exploiting GPU features like large numbers of threads and localized texture memory. We find a surprising variation in the performance that can be achieved on GPUs for our applications and discuss how these findings relate to past known effects in parallel computing such as memory speed-related super-linear speed up.

Parallel graph component labelling with GPUs and CUDA

“…Our experiments have showed that this technique can slightly improve the kernel execution time when processing scale-free graphs with their "fat tail" degree distributions, but the overall execution time, which includes the time required to sort the vertices by the CPU, increases considerably. These findings are in line with previous results reported in [29]. Consequently, the implementation of Kernel 2 reported here does not sort the vertices.…”

“…We can also make use of specialised memory types available on the GPU to improve memory access. In our previous work [29] it was shown that the optimal memory type to use for simulating field equations is the texture memory type. The texture memory type uses an on-chip memory cache that caches values from global memory in the spatial locality.…”

“…The Cahn-Hilliard equation is a phasetransition model of a quenching binary alloy, component labelling can be used to determine which phase the field is currently in. The Cahn-Hilliard simulation used is the GPU-based simulation described in [29]. By testing the labelling algorithms in conjunction with a real GPU simulation, it will provide performance data of the algorithms as they would be used in a real situation.…”

Design and Implementation of GPU-Based Prim's Algorithm

“…Many later studies [17] use that data structure as foundational graph representation. A. Leist et al [18] propose another kind of graph representation which makes some improvement on compact adjacency list. As Fig.…”

Auto-generation of Parallel Finite-Differencing Code for MPI, TBB and CUDA

“…It is possible to bring parallelism to bear on many problems using hybrids of cluster-computing approaches; accelerator technologies such as general purpose graphical processing unit (GP-GPU); and the use of many threads within a conventional multi-core CPU. These are typified by software technologies such as the open standard Message Passing Interface (MPI) [12], [13]; NVIDIA's Compute Unified Device Architecture (CUDA) [14]- [16] for GPUs; and Intel's Thread Building Blocks (TBB) [17], [18] software for multi-threaded programming multi-core devices, respectively. It is however tedious, error prone and non- trivial for a programmer or even a programming team to implement an application that works well across all these three parallel paradigms or platforms -even for a problems like finite-difference equation solving that have relatively well known solutions.…”