The use of GPUs to accelerate scientific workloads is becoming ever more common place. GPUs offer substantially more floating point performance and memory bandwidth than CPUs and with careful programming can offer orders of magnitude improvement in performance without the need for any additional approximations. The use of GPUs in Molecular Dynamics simulations is bringing about a transformation in the timescales that can be accessed without requiring substantial capital investment. I present here recent developments in the acceleration of the AMBER Molecular Dynamics Code PMEMD on NVIDIA Fermi GPUs. Performance on just 8 GPUs is more than twice that achievable for a single calculation on any current conventional supercomputer. The hybrid precision model developed as part of this work provides energy conservation indistinguishable from a full double precision implementation.  Microsecond and longer MD simulations of systems containing tens to hundreds of thousands of atoms is now routine on deskside hardware.

Supercomputer in a desktop: Routine microsecond molecular dynamics simulations of proteins on commodity hardware: Extreme GPU acceleration of AMBER

243rd National Meeting (2012)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.