Output details

Exploiting graphical processing units for data-parallel scientific applications

<12> This article was one of the earliest and detailed investigations of graphical processing unit performance for general purpose scientific applications. An extensive benchmarking of both regular array problems but also of highly irregularly structured problems has provided a much cited guide on exploiting differing forms of GPU memory available on a varied range of devices, even for applications GPUs were not expected to excel at. Our lengthy article includes explicit source code fragments and a detailed commentary on hardware features and how best to exploit GPUs for simulations backed up by a rigorous set of performance measurement experiments.