GPU Toolchains
Controlling a flying robot with realistic brain models requires a very fast execution of the simulations. Our models ran on GPUs for parallel execution, drastically speeding up the execution time compared to conventional CPUs.
In collaboration between the Sheffield and Sussex teams, we developed an interface that combined high-level abstraction via the SpineCreator Graphical User Interface for model creation, and a code-generation framework via GeNN for implementation on the GPUs.
The models defined in SpineCreator were saved in the SpineML XML format. We used XML Translation to turn this model description in C++ code. These files could then be used in GeNN to create GPU code that was optimised for the model and the types of GPUs that we used. This combination provided portability and flexibility, allowing us to test our models on different platforms.
GeNN
GeNN is a GPU enhanced Neuronal Network simulation environment based on NVIDIA CUDA technology.
Learn more on the GeNN website.
SpineCreator
The Spiking Neural Mark-up Language (SpineML) is a declarative XML based model description language for large scale neural network models.
