E-mail Print PDF

Molecular Mechanics Software optimization

O. Michielin


Acceleration of Molecular Mechanics softwares through the use of hardware accelerator boards

T. Schuepbach and O. Michielin in collaboration with the SIB Vital-IT group.  

Speedup for an MD simulation performed on a 10,000 atoms system, with the CHARMM package compiled with G77 v3.4

Molecular dynamics (MD) has grown into an intense field of research up to the point that in term of computer resources, it uses most of the CPU time of high performance computers around the world. But despite the tremendous power increase of those computers in the last decades, scientists still face the problem of speed in their simulations since MD softwares are not all multithreaded and therefore cannot take advantage of such architectures or simply do not scale correctly due to large amount of communicating data. And even though new trends in computer technology is now focusing on multicore architectures to achieve high performance in both computation and communication, MD softwares may remain memory bound. As a consequence, the Molecular Modelling Group has started different projects aiming at accelerating existing molecular dynamics softwares through the use of hardware accelerators boards such as NEC MD Engines, GPUs, FPGAs or Clearspeed boards, to name of few. As of today, the port of the NEC MD Engine has been implemented in the CHARMM software and first results demonstrate a great potential in the field of high cutoff simulation.
In the meantime we have developed several Generalized Born Molecular Volume implementations for GPUs and SSE processors.

google analytics