Simulating Morphologically Detailed Neuronal Networks at Extreme Scale
Ovcharenko Aleksandr, Kumbhar Pramod, Hines Michael, Francesco Cremonesi, Ewart Timoth eacute e, Stuart Yates, Sch uuml rmann Felix, Delalondre Fabien · Advances in parallel computing · 2016
In order to achieve performance at very large scale, the core functionalities of the widely used NEURON simulator have been isolated and optimized into a new application CoreNeuron. CoreNeuron has been developed with the goal of minimizing memory footprint and maximizing scalability. This paper presents the scaling behaviour and performance of CoreNeuron on up to 28 racks of the IBM Blue Gene/Q, with one MPI process per node and 64 OpenMP threads per MPI process, comprising 458,752 cores and 1,835,008 threads at maximum.