A novel CPU/GPU simulation environment for large-scale biologically realistic neural modeling

Area of Science:

• Computational Neuroscience
• Neuroscience
• Computer Science

Background:

• Realistic neural simulations are crucial for studying complex brain structures.
• Increasing biological detail in models leads to longer simulation execution times.
• Existing neural simulators often struggle with performance when detailed biological models are used.

Purpose of the Study:

• To present a novel, efficient simulation environment for large-scale biological neural networks.
• To overcome the performance limitations of current simulators in handling biologically detailed models.
• To introduce the NeoCortical Simulator version 6 (NCS6) as a solution.

Main Methods:

• Development of a parallelizable and scalable CPU/GPU simulation environment (NCS6).
• Implementation of built-in leaky-integrate-and-fire (LIF) and Izhikevich (IZH) neuron models.
• Inclusion of a plug-in interface for custom neuron models.
• Distribution of data across a cluster of eight machines, each with two GPUs.

Main Results:

• NCS6 achieves significant acceleration of neural simulations compared to CPU-only or single-GPU approaches.
• The simulator can handle large-scale networks, simulating one million cells and 100 million synapses.
• Quasi real-time simulation performance is achieved on a distributed cluster.

Conclusions:

• NCS6 offers a powerful, open-source platform for large-scale neural simulations.
• The combined CPU/GPU approach effectively addresses performance bottlenecks in computational neuroscience.
• NCS6 enables more complex and biologically realistic brain simulations efficiently.