implement Potjans-Diesmann model

brian-team / brian2cuda

A brian2 extension to simulate spiking neural networks on GPUs

https://brian2cuda.readthedocs.io/

GNU General Public License v3.0

60 stars 12 forks source link

Open moritzaugustin opened 6 years ago

moritzaugustin commented 6 years ago

https://zenodo.org/record/1244116#.W2xh2519jmE add to speedtests -- if not straightforward just omit this example

moritzaugustin commented 6 years ago

Marcel has reviewed the rescience approach and might know fix possibilites if required (but omit if complicated)

mstimberg commented 6 years ago

I think there are two main reasons why I'd rather not use this model (but please go ahead and try if your feel motivated ;) ):

It has a large number of NeuronGroup/Synapses objects, which already with C++ standalone leads to a long compilation time. I expect it to be worse with CUDA.
It uses a lot of memory, something like 12GB if I remember correctly -- I don't think it will fit on the GPU without some optimization.