Related Experiment Video
Updated: Dec 28, 2025
10:50
Computational Modeling of Retinal Neurons for Visual Prosthesis Research - Fundamental Approaches
Published on: June 21, 2022
2.1K
Understanding Computational Costs of Cellular-Level Brain Tissue Simulations Through Analytical Performance Models
Francesco Cremonesi1, Felix Schürmann2
1Blue Brain Project, Brain Mind Institute, École polytechnique fédérale de Lausanne (EPFL), Campus Biotech, 1202, Geneva, Switzerland.
Neuroinformatics
|February 15, 2020
Summary
Computational neuroscience models face performance bottlenecks. Synaptic modeling, not neuron detail, dictates memory scaling. Future whole-brain simulations require careful co-design of models and hardware.
Area of Science:
- Computational neuroscience
- Neuroscience
- High-performance computing
Background:
- Computational modeling and simulation are crucial for understanding brain complexity and component interactions.
- Diverse brain processes necessitate various model abstractions and specialized tools, often requiring high-performance computing.
- A systematic analysis of computational kernel complexity has been lacking, hindering bottleneck identification.
Purpose of the Study:
- To systematically explore the performance landscape of in silico brain tissue simulations using analytic performance modeling.
- To identify computational bottlenecks in current brain models and project future hardware and software requirements.
- To guide the co-design of brain models and simulation engines for emerging computer architectures.
Main Methods:
- Systematic performance analysis based on analytic modeling.
- Evaluation of three representative in silico models: current-based point neurons, conductance-based point neurons, and conductance-based detailed neurons.
- Characterization of memory bandwidth saturation and shared-memory scaling properties.
Main Results:
- Synaptic modeling formalism (current-based vs. conductance-based) significantly impacts memory bandwidth saturation and scaling, more so than morphological detail.
- Network latency and memory bandwidth emerge as critical bottlenecks for all model abstractions as neuron count increases.
- Current general-purpose computing, while powerful, will face limitations with large-scale neural simulations.
Conclusions:
- The choice of synaptic modeling formalism is a key factor in the performance of neural simulations.
- Future advancements in whole-brain modeling on next-generation computers necessitate addressing network latency and memory bandwidth limitations.
- Co-designing models and hardware is essential for achieving large-scale, efficient brain simulations.
