Fast Simulation of a Multi-Area Spiking Network Model of Macaque Cortex on an MPI-GPU Cluster

Area of Science:

• Computational neuroscience
• Neuroscience
• High-performance computing

Background:

• Spiking neural network (SNN) models are crucial for simulating neuronal population dynamics and brain function.
• Advancements in parallel computing enable large-scale brain simulations with increasing detail.
• GPU-accelerated computing offers new avenues for speeding up complex neural simulations.

Purpose of the Study:

• To evaluate the performance of NEST GPU, a GPU-accelerated library for SNN simulations.
• To assess the efficiency of a novel MPI-based remote spike communication algorithm on GPU clusters.
• To compare NEST GPU performance against the CPU-based NEST simulator for large-scale brain models.

Main Methods:

• Utilized NEST GPU, a CUDA-C/C++ library for SNN simulations on GPUs.
• Implemented a novel MPI-based algorithm for remote spike communication on a GPU cluster.
• Simulated a 32 mm² multi-area model of macaque vision-related cortex (approx. 4 million neurons, 24 billion synapses).
• Compared simulation results and performance metrics against the CPU-based NEST simulator on a high-performance computing cluster.

Main Results:

• NEST GPU simulations showed optimal statistical agreement with NEST across three key neural activity distributions.
• NEST GPU achieved significant speed-ups in simulation time per second of biological activity.
• Simulating the metastable state of the macaque cortex model was 3.1x faster with NEST GPU compared to NEST.
• Simulating the ground state of the macaque cortex model was 2.4x faster with NEST GPU compared to NEST.

Conclusions:

• NEST GPU provides a highly efficient platform for large-scale SNN simulations.
• The novel remote spike communication algorithm enhances performance on GPU clusters.
• GPU acceleration offers substantial speed-up for complex, biologically realistic neural network models.