This proposal is for adding MPI support to GeNN. GeNN is known as a GPU-enhanced Neuronal Network simulation environment based on code generation for Nvidia CUDA. However, the limitation of GeNN is that it can only support running on a single GPU or a single shared-memory machine. This proposal is to plan for expanding GeNN to multiple machine clusters. MPI is common message passing interface and infrastructure to communicate across multiple hosts and it is natural to expand GeNN with MPI interface to achieve parallel execution on multiple hosts. The key point of this project is to balance GeNN computation simulations across MPI hierarchy between hosts and block/thread on individual GPU. Furthermore, This proposal shows potential tuning directions basing on MPI-CUDA hierarchy. To bridge the gap between the destination and implementation, this proposal also describes feasible stages for the MPI-GeNN project and narrates the candidate’s motivation, background and progressive approach. Curriculum vitae is attached to the last pages for reference. The candidate is welcome to any questions, comments or suggestions through any of the contacts, such as email, telephone, and skype.

Organization

Student

Mengchi Zhang

Mentors

  • Jamie Knight
  • Thomas Nowotny
close

2017