Folding Forum

I have been playing around with 5-6 MD programs and I would have to say by far the best has been OpenMM Zephyr. Installation was extremely easy and even an idiot like me can work it!.

Anyways, when I was playing around with that program with gpu acceleration, whenever I used implicit solvent my gpu usage of a gts 250 was 2% for Sacolipin, and 44% for Titin.

However when I used explicit for either proteins the gpu usage was 99%. I assume that OpenMM Zephyr works very, if not that same as f@h because it uses GROMACS and is developed by Stanford. Edit: So since f@h gpu uses 99% gpu I assume f@h uses explicit.

I know implicit solvent is much quicker (but less accurate?) and was wondering if that is the reason you guys use explicit simulations because it is more scientifically accurate?

Thanks for your time for reading and your answer

Have you seen this page: https://simtk.org/home/openmm?
F@h uses a combination of explicit and implicit solvation models. Both have advantages and disadvantages. In general for Folding@home: GPUs perform implicit solvation models, CPUs perform explicit, and PS3 does mostly implicit the fastest although it can also do explicit if need be. It is my understanding that implicit and explicit models just have different approaches, and reveal different things, such as how water affects protein folding (http://www.pnas.org/content/101/17/6456.full)

There are certainly more knowledgeable people here who can give you a better answer, but I wanted to get some info together for you. OpenMM is simply an abstraction layer, but GPUs perform implicit solvation because they're fast at that.

EDIT: I have been informed by an admin that implicit solvation models need less processing because there is intrinsically less work to do. They lead to approximate answers, which may or may not be sufficient to bring about scientific conclusions. If more accuracy is required, the results can be refined by creating explicit models based on what has been learned from the implicit models and focusing on the details that need to be refined. The choice of implicit and explicit depends on a desire for this accuracy, and available resources. Since F@h supports a variety of different hardware, each with unique architectures and capabilities, the Pande Group can use both models.

I'm no expert, however it is my assumption that GPU's use implicit based on this evidence...

If you look on the project table http://fah-web.stanford.edu/psummary.html you see that the OpenMM units typically have around 600 atoms, this would suggest that water is an implicit model and not included in the "atom count".

as opposed to the Gromacs ones where it can be anywahere up to 2.5m atoms.

Regards

Matt