Video-Gaming Hardware for Scientific Supercomputing

This section is dedicated to the ongoing research projects of our group related to the use of video-gaming technologies for scientific computation. This work is currently supported under NSF grant PHY-01414440 and AFRL CRADA agreement 10-RI-CRADA-09. Initials of the faculty involved, are in parentheses. Here is a list of research articles published using results generated from this effort: Preprint arXiv:1312.5210 (2013)Phys. Rev. D90 084025 (2014)CSC’14 (2014)Phys. Rev. D91 104017 (2015)Phys. Rev. D93 041501R (2016)Phys. Rev. D94 084049 (2016)IEEE HPEC Conf. (2017)Phys. Rev. D95 081501 (2017)Phys. Rev. D96 024020 (2017)Class. Quant. Grav.  34  205012 (2017);


The Sony PlayStation 3 Gravity Grid (Coyne, Khanna)

This NSF supported project has its own dedicated website

Video-Gaming Technologies for Scientific Computing in Gravitational Physics (Coyne, Khanna)

This NSF supported project began in 2014. We plan to explore the capabilities of current and next generation consumer-grade, video-gaming hardware for numerical modeling and big data analysis in the area of gravitational wave science. Specific examples of the compute hardware that we consider interesting for this study are current generation gaming GPUs like the AMD Radeon HD & Nvidia GeForce series and also the CPU-GPU hybrid or “fused” processor architectures like AMD Fusion APU, Intel’s Ivy Bridge, ARM SoC and next generation gaming consoles such as the Sony PS4.

The main advantage of considering such consumer-grade hardware for scientific computing is its very low cost and high power-efficiency. The parallel software development framework that we utilize in this work is the Open Computing Language (OpenCL).

Partial results of this exploration are available in this presentation made at the AFRL, Rome NY in November 2014: KhannaCRADA2014.pdf and the standardized benchmarks of various compute kernels: SHOC

The final 2017 results of an extensive study in this context are available in this IEEE HPEC Conf. (2017) publication. 

Currently, the most promising approach towards developing a very low-cost clustered configuration that is highly power efficient  as well is to use a streaming media type device as a cluster node. An excellent candidate for such a system is a cluster built using Nvidia Shield TV media players. A small solar-powered prototype system using these was released on HPC Day 2017. DIY instructions may be found here: ShieldDIY.