Talk, Ada Sedova
Nov 28, 2016
from 11:00 AM to 12:30 PM
|Where||2040, third floor classroom|
|Contact Name||Lora Davis|
|Add event to calendar||
Acceleration, sampling, portability and force field accuracy: prospects and challenges for MD as we move past the petascale
Recent accelerator architectures have allowed us to simulate unprecedented system sizes and approach experimental timescales. Results from these simulations, however, have sometimes pointed strongly to model inaccuracies that previously could be ascribed to inadequate sampling due to computational limitations. Moreover, with the creation of multiple types of commercially available accelerators comes the problem of creating portable programs that do not have to be rewritten for every machine type. This problem is a difficult one to tackle as each new architecture can require the use of a new programming language and unique memory considerations. On the other hand, with each new development in computing acceleration comes the potential for revisitation of approximations that were previously made due to computational limitations, and thus to the possibility of developing a more accurate model. Both sides of the the story relating to next-generation molecular dynamics will be presented.