Implementation of Izhikevich spiking neurons running on the CPU and in compute shaders on the GPU using wgpu-rs. Mostly an excuse for trying out compute shaders and GPU programming.

Milestones

0.2

Replicate the setup from 0.1 but make it run and graph continuously.

Currently working but graphing the individual spikes is extremely slow. I think it's more due to using the graphing library in ways it wasn't meant to be used.

The no-spikes flag will only draw neuron 0's voltage over time which is significantly faster.

0.1

A replication of the example code in the the paper that produces a similar graph output.

Building

Should Just Work™ but I found on Windows, compiling shaderc from source used up computer-breaking amounts of RAM so I had to use the prebuilt lib and the SHADERC_LIB_DIR envvar option.

Running

Default is running on the GPU

cargo run -- 1000

To use the CPU:

cargo run -- --cpu 1000

The resulting graph defaults to ./out.png but can be changed.

Increasing the number of neurons increases RAM usage exponentially due to the dense connection matrix.