Gpu acceleration cuts latency and energy for neuron core mapping
GPU-Accelerated Hypergraph Partitioning and Placement to Map SNNs on Neuromorphic Hardware
Distributed, Parallel, and Cluster Computing
Summary
Spiking neural networks use bursts of activity to communicate efficiently across many small processors. How these neurons are assigned to the processors affects how fast and energy-efficient the system runs. The authors created a way to speed up this assignment on graphics processors by breaking the problem into pieces and improving connections between processors. Their approach reduces the time data spends moving around and uses less energy compared to older methods, while running much faster.
What this means in practice
- •For neuromorphic hardware engineers: Improve spike communication speed and reduce energy use by efficiently assigning neurons to hardware cores using a GPU-accelerated mapping pipeline.
- •For high-performance computing teams: Use parallelized hypergraph partitioning to accelerate large-scale graph partitioning tasks on GPUs beyond neuromorphic contexts.
Authors
Marco Ronzani, Cristina Silvano
Abstract
SNNs running on neuromorphic hardware use spikes to achieve sparse and energy-efficient communication over a mesh of cores. In turn, system performance heavily depends on the assignment of neurons to cores: the mapping. Since hardware features inter-core multicast and intra-core replication of spikes, we model SNNs as hypergraphs to exploit both opportunities for reducing communication traffic. Mapping thus comprises two NP-hard problems: hypergraph partitioning and placement on the lattice of cores. High-quality solutions to both are critical, yet increasingly difficult as networks scale to millions of neurons. Therefore, we propose a GPU-accelerated pipeline for SNN mapping: a multi-level partitioning scheme is devised around hardware constraints, while placement is initialized through recursive bisection, followed by refinement pulling together strongly connected cores through repeated swaps. Model-based experiments show upwards of 16% lower latency and 42% lower energy for spike movements over existing sequential tools, while our parallel mapper is on average 18-280x faster.