Complex, large scale neural systems are difficult to construct and tune. Modelers must manipulate vast quantities of data to make progress. The Iterative Evolution of Models (ItEM) system is designed to automate and simplify as much of this work as possible to increase the cadence of model development.
Reproducible science and scalability are secondary goals of the ItEM system. By automating as much as possible, ItEM allows researchers to more easily reconstruct their own past results. ItEM also stores data and manages cluster computing resources centrally, so modeling work can scale as quickly as additional hardware can be brought online.
Initially, ItEM will run large-scale neural models written for Cog Ex Machina on a new GPU cluster hosted at HP Labs under the direction of Greg Snider and Dick Carter. The HP cluster, called Simcity, features a total of 144 GPUs, 576 GB of conventional memory, 432 GB of GPU memory, and an Infiniband interconnect. A prototype cluster containing three nodes and six GPUs, called Simtown, is also available for testing and debugging. By the end of 2011, the Neuromorphics Lab will finish building out a twin system with roughly half the computing power of Simcity.