MC steps and MD simulation time
-
- Posts:3
- Joined:Wed Jul 23, 2008 7:20 pm [phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1275: count(): Parameter must be an array or an object that implements Countable
How many MC steps corresponds to 1 ns MD for a protein with 300 residues?
Re: MC steps and MD simulation time
That's a very good question. I don't think it has a simple answer.... I am looking forward to further replies
.
![Smile :-)](./images/smilies/icon_e_smile.gif)
-
- Developer
- Posts:38
- Joined:Sun Jun 15, 2008 10:27 pm [phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1275: count(): Parameter must be an array or an object that implements Countable
Re: MC steps and MD simulation time
We have never tested a system of this size, so it's difficult to answer.
For smaller systems (20-30 residues) we found that one MC scan (i.e. the number of MC moves required to move every residue once) equals to roughly one 2 fs MD time step.
For smaller systems (20-30 residues) we found that one MC scan (i.e. the number of MC moves required to move every residue once) equals to roughly one 2 fs MD time step.
benchmarking (Re: MC steps and MD simulation time)
I have a system of 10400 atoms (1293 residues).
I am running Monte Carlo (MC) with implicit membrane with hippo [with Supplemental Streaming SIMD Extension 3 (SSSE3)]. For this, a scan is about 1e4 moves; let’s say this corresponds to a 2 fs molecular dynamics (MD) time-step. The fastest I can get on a Intel Xeon CPU L5410 2.33 GHz processor was 20 moves per second. Running 1e7 steps (? 1e3 scans ? 2 ps) takes ?1 week. (1 ns would therefore take 500 weeks ? 10 years.)
By contrast, I can run an MD simulation with Gromacs of this same system in vacuum on an Intel Xeon CPU 1.7 GHz processor and get 1 ns in 1 day.
Am I being fair here? I wonder whether it is possible to implement parallel MD with implicit membranes, and how fast would that be?
I am running Monte Carlo (MC) with implicit membrane with hippo [with Supplemental Streaming SIMD Extension 3 (SSSE3)]. For this, a scan is about 1e4 moves; let’s say this corresponds to a 2 fs molecular dynamics (MD) time-step. The fastest I can get on a Intel Xeon CPU L5410 2.33 GHz processor was 20 moves per second. Running 1e7 steps (? 1e3 scans ? 2 ps) takes ?1 week. (1 ns would therefore take 500 weeks ? 10 years.)
By contrast, I can run an MD simulation with Gromacs of this same system in vacuum on an Intel Xeon CPU 1.7 GHz processor and get 1 ns in 1 day.
Am I being fair here? I wonder whether it is possible to implement parallel MD with implicit membranes, and how fast would that be?
-
- Developer
- Posts:38
- Joined:Sun Jun 15, 2008 10:27 pm [phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1275: count(): Parameter must be an array or an object that implements Countable
Re: MC steps and MD simulation time
This sounds very slow. What cutoff are you using? With a 14 Angstrom cutoff you should get at least 1000 moves/s on the Xeon.
The MD version of the membrane is implemented and should be much more efficient for larger systems. It still needs a little debugging and probably some form of stochastic dynamics. I will try and do this asap.
Definitely before X-mas!
The MD version of the membrane is implemented and should be much more efficient for larger systems. It still needs a little debugging and probably some form of stochastic dynamics. I will try and do this asap.
![Wink ;)](./images/smilies/icon_e_wink.gif)
Re: MC steps and MD simulation time
Thanks! I changed the cutoff from 44 Å to 14 Å and am getting about 200 moves per second.