[Comp-neuro] Course on parallelizing NEURON models

Ted Carnevale ted.carnevale at yale.edu
Mon Apr 1 04:34:38 CEST 2013

Who should take the course on parallelizing NEURON models that
we are planning to present June 26-30 at the Institute for Neural
Computation at UCSD?  Any NEURON user who:

* wants to use all of the computational power of their own multicore
   laptop or desktop PC or Mac, in order to run simulations more
   quickly--gain an N-fold speedup on an N-core machine!

* has a model optimization or parameter space exploration problem
   that requires hundreds or thousands of time-consuming simulations

* has a model that is too complex for a single PC or Mac, and wants
   to take advantage of high performance parallel computing resources

This course will address topics that include:
* installing NEURON on a wide variety of parallel supercomputers
* parallelizing network models that involve spike-triggered synaptic
   transmission and/or gap junctions
* using "bulletin board style parallelization" for embarassingly
   parallel problems such as parameter space exploration
* ensuring that simulation results are independent of the number of
   processors or how cells are distributed over the processors
* implementing reproducible randomness (and why it is essential)
* measure performance
* achieving load balance (a key factor in maximizing performance)
* debugging parallel models
* using remote high-performance computing resources

Applicants who sign up and pay by Friday, April 12, will qualify for
the early bird registration fee of $1200.  After April 12, the fee
goes back up to $1350.  Registration is limited, and the last day
to sign up is Friday, May 24.  For more information and the on-line
application form, see
or contact
ted dot carnevale at yale dot edu

More information about the Comp-neuro mailing list