A solver for sets of linear systems for neural network simulations in CUDA

• Grant number: 13/14603-9
• Support Opportunities: Scholarships in Brazil - Master
• Start date: October 01, 2013
• End date: September 30, 2014
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Raphael Yokoingawa de Camargo
• Grantee: Saeed Shariati
• Host Institution: Centro de Matemática, Computação e Cognição (CMCC). Universidade Federal do ABC (UFABC). Ministério da Educação (Brasil). Santo André , SP, Brazil

Abstract

Nowadays, the general-purpose computing on graphical processing units (GPGPU) technique is widely accepted as a new paradigm of High Performance Computing for improving the performance of simulations, such as large-scale neuronal simulations. In this work we will develop an algorithm for solving a large sets of linear systems, each representing a single neuron, using GPUs. We will incorporate this algorithm into the MOOSE neuronal simulator, which is the successor of the widely used GENESIS simulator. We will then compare the accuracy and performance of the GPU and CPU versions of the simulator. With this work, we expect to contribute to the area of neuroscience, by providing a more efficient simulator for the community of neuroscientists, and to the area of high-performance computing, by developing a GPU algorithm that will be useful for several kinds of simulations.