I need to run direct frequency response (SOL108) for a model with 3.5M dof on a Windows workstation with dual 8-core processors, 256 Gb RAM and 1Tb SSD. Using SMP and parallel=8 it takes about 12 hours to compute 750 frequency points. I'm looking for ways to speed p the solution significantly. I would like to know if running this problem using DMP on the same machine with 2 or 4 cores would be any faster than SMP. Or would the I/O overhead with a single SSD outweight any performance advantage of DMP. Your collective experience is requested.