Relative performance of various GPUs

The functionality of OpenMM will (eventually) include everything that one would need to run modern molecular simulation.
POST REPLY
User avatar
Peter Eastman
Posts: 2553
Joined: Thu Aug 09, 2007 1:25 pm

Re: Relative performance of various GPUs

Post by Peter Eastman » Thu Apr 09, 2015 9:57 am

No, because AMBER is using 4th order b-splines while OpenMM is using 5th order. That makes the AMBER results much less accurate.

Peter
Top
User avatar
Mark Williamson
Posts: 31
Joined: Tue Feb 10, 2009 5:06 pm

Re: Relative performance of various GPUs

Post by Mark Williamson » Sun Apr 19, 2015 5:51 am

Hi Peter,
Apologies for delay in response. The AMBER CUDA implementation does not support interpolation orders greater than 4, and OpenMM's order parameter seems not to be user definable. Hence this comparison cannot be made.

Thank you for the clarification.

Mark
Top
User avatar
Peter Eastman
Posts: 2553
Joined: Thu Aug 09, 2007 1:25 pm

Re: Relative performance of various GPUs

Post by Peter Eastman » Sun Apr 19, 2015 9:15 am

The comparison definitely can be made. But you need to actually measure the accuracy of the results. Even if you could use identical parameters in both programs, that wouldn't be a meaningful comparison: the parameters that are optimal in one program might not be optimal in the other. To compare them, you need to measure the accuracy of the results in both programs, then tune the parameters in each program to get optimal performance while achieving equal accuracy.

Peter
Top
POST REPLY