CoreNEURON : An Optimized Compute Engine for the NEURON Simulator

Abstract

The NEURON simulator has been developed over the past three decades and is widely used by neuroscientists to model the electrical activity of neuronal networks. Large network simulation projects using NEURON have supercomputer allocations that individually measure in the millions of core hours. Supercomputer centers are transitioning to next generation architectures and the work accomplished per core hour for these simulations could be improved by an order of magnitude if NEURON was able to better utilize those new hardware capabilities. In order to adapt NEURON to evolving computer architectures, the compute engine of the NEURON simulator has been extracted and has been optimized as a library called CoreNEURON. This paper presents the design, implementation, and optimizations of CoreNEURON. We describe how CoreNEURON can be used as a library with NEURON and then compare performance of different network models on multiple architectures including IBM BlueGene/Q, Intel Skylake, Intel MIC and NVIDIA GPU. We show how CoreNEURON can simulate existing NEURON network models with 4–7x less memory usage and 2–7x less execution time while maintaining binary result compatibility with NEURON.

Understanding Computational Costs of Cellular-Level Brain Tissue Simulations Through Analytical Performance Models

“…Several GPU implementations of brain tissue simulations have been proposed in the literature (Brette and Goodman 2012;Fidjeland et al 2009;Yavuz et al 2016;Kumbhar et al 2019b). A comparison between GPUs, HPC hardware and neuromorphic hardware found that, under certain conditions, GPUs can beat neuromorphic hardware in terms of energy efficiency but not an HPC server in terms of performance (Knight and Nowotny 2018).…”

“…All the benchmarks were executed multiple times under the same conditions (typically around 10 runs), and we define the error (or margin of error) as the ratio of the difference between the median measurement and predicted runtime divided by the median measurement. For validation and benchmarking we use the CoreNEURON implementation as reference (Kumbhar et al 2019b).…”

Dense Computer Replica of Cortical Microcircuits Unravels Cellular Underpinnings of Auditory Surprise Response

“…The densities of ionchannels on morphologically-detailed neuron models were optimized to reproduce the behavior of different electrical neuron types (e-types) and synaptic dynamics recorded in vitro 56 . Simulations were run on HPE SGI 8600 supercomputer (BlueBrain V) using NEURON 57 and CoreNEURON 58 . NEURON models and the connectome are available online at bbp.epfl.ch/nmc-portal 59 .…”

Exploring Complex Brain-Simulation Workloads on Multi-GPU Deployments

“…NEURON supports multiple neuron compartments called sections (and referred to as unbranched cables). CoreNEURON is a simplified engine for the NEURON simulator optimized for both memory usage and computational speed [17]. Its goal is to simulate larger network models on modern supercomputing platforms and to reduce memory footprints.…”

“…GPUDirect [25] is a technology that permits the disaggregation of (GPUDirect-enabled) GPUs from CPUs and enables a direct path for data exchange among GPUs on, for instance, independent compute nodes in a cluster, with the purpose of keeping the latency to a minimum. Then, true multi-GPU simulations whereby inter-GPU data exchange occurs with minimal-latency technology, instead of having to stage the data exchange by various memory copies, of which [2,17] are examples. For instance, in molecular dynamics, the Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) [27] or the Amber Package [28] are software for performing large-scale trajectory simulations, which GPUDirect has reduced from days to hours.…”