FDS handles parallelism at the mesh level. Each iteration, every mesh in your model is independently updated, then the values on the boundary of each mesh are communicated between the different processes and the simulation continues. You can run a 1,2, 4, or 37 mesh problem on a machine with 4 computing cores, but the 1 mesh version would only use 1 computing core and the 37 mesh version would use all 4 cores but suffer from communication overhead when sending the boundary data back and forth. You get the fastest result by matching the number meshes to available computing cores.
PyroSim will run quite well on multiple cores. In the office it is running on single CPU, dual CPU, dual core, and quad core machines with no issues. We do not currently have a version of PyroSim that takes advantage of the increased memory potentially available in 64-bit architectures. In order to run a single process (i.e. mesh) that uses over 3 GB of memory, you will need to download the 64-bit version of FDS from NIST. Then use the Preferences dialog on the File menu, to point PyroSim to the new executable.
We do not maintain a spec for the simulation machine, but the topic is discussed on the FDS users forum. You may find it useful to read through some of the discussions there. This is a good place to start:
http://groups.google.com/group/fds-smv/browse_thread/thread/bcafaa5cad36c06e
