Year: 2016
Type: article
Abstract: A Kampmann-Wagner type numerical precipitation model (KWN) has been implemented using NVIDIA’s CUDA framework for numerical programming of the graphics processing unit (GPU). Different implementation strategies are discussed and subjected to performance When the KWN model is used in combination with other calculations that are processed by the CPU, the performance improvements can be such that the KWN model incurs only emph{negligible} additional execution time. Also if the KWN model is used standalone for a large case, the GPU implementation achieves good scalability and performance.measurements. We study two representative cases corresponding to a large and a small workload. The model is found to be well suited for a GPU implementation, provided that there is enough work to keep the device busy and the right parallelization strategy is chosen. For our hardware, we recommend a minimum work load of more than $2^{16}$ histogram bins (as the total of multiple histograms) which corresponds to 146 histogram bins per GPU core.
Authors: Øyvind Jensen, Henrik Lam, Hallvard Fjær
Institutions: Institute for Energy Technology, IMPETUS Afea (Sweden)
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid:
Funders: Nvidia, Iron and Steel Institute of Japan
OpenAlex ID: W4385786063