Re: [AMBER] Implicit solvent simulation is unexpectedly slow

From: David A Case <>
Date: Sun, 2 Aug 2015 08:30:43 -0400

On Sun, Aug 02, 2015, 浅井 賢 wrote:
> Several years ago, I performed a simulation with explicit solvent model,

> Recently, I performed a similar simulation with implicit solvent model,

You can't expect to compare the timings of these two sorts of simulation
models in any obvious fashion. The cuda implementation of GB does not
provide for cutoffs, so you are explicitly evaluating all pair interactions.
By contrast, explicit solvent simulations use PME to handle long-range

The Amber GPU benchmark page has lots of good data. It shows that a
25,000 atom system with GB runs at something like 7 ns/day on two very
recent GPUs. This doesn't sound all that different from what you report.
That page also shows a 90,000 atom explicit solvent system that runs at
50ns/day, for comparison. Explicit solvent calculations are often much faster
than the correpsonding implicit ones.

If you are worried about the settings you are using, compare them to those
given the benchmark page.

....hope this helps....dac

AMBER mailing list
Received on Sun Aug 02 2015 - 06:00:04 PDT
Custom Search