Hi Anu,
Think of this as a standard Disclaimer rather than a flat out warning. Essentially the CPU code has been around for many years and has been widely tested. The GPU code on the other hand is very new.
The key here is that if you see strange behavior from the GPU code - by strange I means simulations crashing etc - try it with the CPU code and see if the problem persists. If it does not then write to the AMBER list with details of what you are trying to simulate. The GPU code is now heavily used - upwards of 400 citations and I would guess something like 80% of AMBER PMEMD simulations are now likely with GPUs. So you shouldn't need to worry. Just be aware that if you are simulating some kind of exotic system - as opposed to regular proteins, DNA, RNA and standard organic ligands - the GPU code may give results that don't make sense. It may crash with limited error messages or give give some other result that you cannot rationalize / understand. In this case don't just blindly go down that path trying to make conclusions - consider that it could be coming from a bug in the code.
As with everything in science you should always be devising ways to test your hypothesis in multiple independent ways from different directions. If two different approaches lead to the same conclusion that adds more weight to that conclusion.
I hope that helps. I'll tone down this message for the next release of AMBER since the GPU is much more widely used now. (Being in the US I am obligated to state however that there is still no warranty implied or otherwise or any other guarantee that it is for any purpose however).
All the best
Ross
> On Feb 26, 2016, at 3:28 AM, anu chandra <anu80125.gmail.com> wrote:
>
> Dear Amber users,
>
> I have recently started using pmemd.cuda.MPI (Amber 14) for carrying out MD
> simulations. I have noticed in the .out file that a caution message has
> appeared about the CUDA version of PMEMD. It says...
>
> CAUTION: The CUDA code is currently experimental.
> You use it at your own risk. Be sure to
> check ALL results carefully.
>
>
> I just wonder how should I make sure the iterations are going fine?.
> Precisely, what all results I should check to make sure the simulation goes
> on the right direction?
>
>
> Many thanks in advance
>
> Anu
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Feb 26 2016 - 15:00:05 PST
