A multicore GPU server for modeling hydraulic fractures

点击翻译按钮获取中文摘要

This proposal aims to deploy a parallel architecture workstation based on multiple multicore Graphical Processor Units (GPUs) for the numerical modeling of hydraulic fractures. This supercomputer constructed from consumer GPUs manufactured by nVIDIA will comprise 1792 GPU cores, 8 CPU cores, and will have 72 GB of system memory. The objective of this proposal is to develop specialized algorithms that will exploit the multicore architecture via the CUDA programming language that will enable the rapid modeling of propagating hydraulic fractures. Hydraulic fractures are of great industrial importance in the Oil, Gas, and Mining Industries and are being investigated as a key component in the sequestration of CO2 in impermeable rock. It is expected that this project will provide an ideal training opportunity for HQP in the specialized techniques required to exploit this new form of parallel architecture. It is anticipated that these techniques will also find application in other areas of scientific computation, which share the same challenges as those encountered in the strongly coupled system of equations governing the propagation of hydraulic fractures.