Interactive flow simulation using Tegra-powered mobile devices. (January 2018)

• Record Type: Journal Article
• Authors: Harwood, Adrian R.G.; Revell, Alistair J.
• Is Part Of: Advances in engineering software. Volume 115(2018)
• Page Start: 363
• Page End: 373
• Publication Date: 2018-01
• Subjects: Android -- Mobile computing -- Interactive simulation -- Lattice Boltzmann Method -- CUDA -- Embedded computing; Computer-aided engineering -- Periodicals; Engineering -- Computer programs -- Periodicals; Engineering -- Software -- Periodicals; Periodicals; 620.0028553
• DOI: 10.1016/j.advengsoft.2017.10.005
• Languages: English
• ISSNs: 0965-9978

Abstract

Highlights: A general framework for implementing interactive simulations on mobile devices is formulated. A GPU-powered implementation of the framework is tested and interactivity metrics quantified. Compared with an earlier CPU implementation, the computational efficiency improves by a factor of 300. We find that problems should not be sized for maximum throughput as is typical for GPU-accelerated CFD. Rather, a more suitable metric is detailed based on the perceived rate of flow through the domain. Abstract: The ability to perform interactive CFD simulations on mobile devices allows the development of portable, affordable simulation tools that can have a significant impact in engineering design as well as teaching and learning. This work extends existing work in the area by developing and implementing a GPU-accelerated, interactive simulation framework suitable for mobile devices. The accuracy, throughput, memory usage and battery consumption of the application is established for a range of problem sizes. The current GPU implementation is found to be over 300 × more efficient in terms of combined throughput and power consumption than a comparable CPU implementation. The usability of the simulation is examined through a new 'interactivity' metric which identifies the ratio of simulated convection to real world convection of the same problem. This real-time ratio illustrates that large resolutions may increase throughput through parallelisation on the GPU but this only partially offsets the decrease in simulated flow rate due to the necessary shrinking of the time step in the solver with increasing resolution. Therefore, targeting higher throughput configurations of GPU-solvers offer little additional benefit for interactive applications due to ultimately simulations evolving at a too slow a rate to facilitate interaction. The trade-off between accuracy, speed and power consumption are explored with the choice of problem resolution ultimately being characterised by a desired accuracy, flow speed and endurance of a given simulation. With current rates of growth in mobile compute power expected to continue, real-time simulation is expected to be possible at higher resolutions with a reduced energy footprint in the near future.