Atom Limit

classic Classic list List threaded Threaded
9 messages Options
Reply | Threaded
Open this post in threaded view
|

Atom Limit

Mfennimore
Greetings,

What is the atom limit in abalone. Because my interest is primarily in aqueous surfactant aggregates, I need to be able to build solvation boxes with at least 5,000 to 6,000 water molecules. Currently, I can perform reasonably fast calculations with something like 3500 water molecules; but beyond that I get an error message followed by a crash. Is there anyway to push such calculations though?

Also, I calculated the radial distribution functions of fspc water (Amber 94) and TIP4F water (Amber 99B) both under ambient and super critical conditions (i.e. 673F and 0.64g/cm^3) and the results of the latter were quite excellent; however, the OH radial distribution at ambient conditions was slightly overestimating the first maxima using both force fields. Have you performed these calculations? Did you get a similar result?

Reply | Threaded
Open this post in threaded view
|

Re: Atom Limit

agilemolecule
Administrator
Mfennimore wrote
What is the atom limit in abalone. Because my interest is primarily in aqueous surfactant aggregates, I need to be able to build solvation boxes with at least 5,000 to 6,000 water molecules.
In principle, it can be a very large number of atoms, if disable the use of a "list of atoms".
However, this may slow down the calculation considerably and does not work on the GPU.
In practice it is not used.

On my computer with "list of atoms", get used to about 2000 water molecules, in the mode of "all with all". If limit the distance of the forces action (Rm) that  a higher number can be used.

However, I hope that the next version will handle memory more gently and allow to work effectively with much larger models. I hope that it will be ready soon. Please be in touch.

Mfennimore wrote
Also, I calculated the radial distribution functions of fspc water (Amber 94) and TIP4F water (Amber 99B) both under ambient and super critical conditions (i.e. 673F and 0.64g/cm^3) and the results of the latter were quite excellent; however, the OH radial distribution at ambient conditions was slightly overestimating the first maxima using both force fields. Have you performed these calculations? Did you get a similar result?
Overestimation of the height of the first peak is typical for many models of water.
Reply | Threaded
Open this post in threaded view
|

Re: Atom Limit

agilemolecule
Administrator
This post was updated on .
In reply to this post by Mfennimore
Please try version 1.8.88.
Water benchmarking on http://www.biomolecular-modeling.com/Abalone/Benchmarking.html

Happy New Year!




























Reply | Threaded
Open this post in threaded view
|

Re: Atom Limit

mfennim
This is very interesting; however, I haven't been able to utilized GPU computing on large clusters--though I have a GeForce GTX 760. For example, simulations on 1000 water molecules works fine, but simulations on 3500 water molecules crashes with GPU acceleration. Why is that?

Also, when I perform replica exchange, all of my 4 cores are utilized, but when I do standard MC or dynamics only one core (~20% cpu) is being utilized. Is there a way I can access all cores and GPU for large and small systems alike?

Thanks for the excellent support
Mark
Reply | Threaded
Open this post in threaded view
|

Re: Atom Limit

agilemolecule
Administrator
mfennim wrote
This is very interesting; however, I haven't been able to utilized GPU computing on large clusters--though I have a GeForce GTX 760. For example, simulations on 1000 water molecules works fine, but simulations on 3500 water molecules crashes with GPU acceleration. Why is that?
This is strange. I work with the model of 8000 molecules in all with all mode (GeForce GTX 670).
Whether you are talking about the Toukan-Rahman water model? The fact that it is written a special code for it. Everything else works as before.

For 4-site models you are still limited.
However, you can do the trick. Redefine the force field.
In the file FineType.mlm you can change the types O4F and H4F to OW and HW. The program determines the model according to these parameters. In general, any triple sequence of atoms of OW, HW, HW types considered by this new code. (Of course in this case, you need to adjust the the Lennard-Jones parameters.) L4F will continue to be calculated by the old code.


mfennim wrote
Also, when I perform replica exchange, all of my 4 cores are utilized, but when I do standard MC or dynamics only one core (~20% cpu) is being utilized. Is there a way I can access all cores and GPU for large and small systems alike?
Unfortunately, you can not use GPU and a lot of CPU at the same time.
The Replicas Exchange parallelized using the processors, but the dynamics and Monte Carlo using the GPU. To me it seems rational because the larger models can not be calculated by the Replicas Exchange, not even enough power of the card. A small model works better on multiple processor cores.
If you are interested in thermodynamic properties, I would recommend Hybrid Monte Carlo.




Reply | Threaded
Open this post in threaded view
|

Re: Atom Limit

mfennim
I set up new calculations and the GPU option appears to work great. I'm running simulation on large clusters at least twice as fast compared to non GPU computing. I'm not sure what the issue was, but it's been resolved.

Thanks again. Happy Newyears!
Mark
Reply | Threaded
Open this post in threaded view
|

Re: Atom Limit

agilemolecule
Administrator
The most likely matter is the default integrator. If you are using a multi-step integrator, then intermediate forces come first, which are calculated on CPU. It might make sense to switch to the Verlet integrator. It can not do more than 0.5-1 fs steps, but it is good for acceleration on GPU.
DR
Reply | Threaded
Open this post in threaded view
|

Re: Atom Limit

DR
In reply to this post by agilemolecule
Hi there,
I am trying to run the following system:
# System
Molecule_types         2
~C43H41NS              38
H2O                    19641
PBC                rect
Box                87.99  87.997  87.993
And I get the following error message in the output file:

!!! list of far neigbours overfilled 0
Insrease NBLMAX in prcm.h for this job to  910568564

Is there a windows compiled version that can handle this system (or bigger as I still need to add a fixed wall of silica)? If not, any other options to solve this issue?
Thanks!
Reply | Threaded
Open this post in threaded view
|

Re: Atom Limit

agilemolecule
MDynaMix was compiled with g77 http://kkourakis.tripod.com
You can try changing the size of the arrays to your task.