Re: [AMBER] Brand new GPU card testing

From: Ross Walker <ross.rosswalker.co.uk>
Date: Thu, 19 Apr 2018 08:11:50 -0400

Hi Nikolay,

I suggest running the same calculation, with the same random seed, repeatedly for about 24 hours and check that you get identical results in each case. I typically run 3 sets of calculations. One around 25000 atoms, one around 1 million atoms and one around 3.5 million atoms. The 3.5 million atom calculation is designed to use the majority of the cards memory but it isn't as simple as just testing all the memory, the actual workload matters. I've seen cases where the 3.5 million atom calculation works fine but the 25000 atom run gives errors.

I typically set nstlim such that each calculation takes about 30 mins - so 50 or so repeats gives you ~ 24 hours.

Hope that helps.
All the best
Ross

> On Apr 19, 2018, at 03:28, Nikolay N. Kuzmich <nnkuzmich.gmail.com> wrote:
>
> Dear Amber users and developers,
>
> how a brand new graphic card (1080Ti GTX, for instance)
> should be tested to make sure it's quite suitable for running the
> simulations?
>
> Kind regards,
> Nick
>
> Nikolay Kuzmich
> Department of Drug Safety,
> Research Institute of Influenza,
> WHO National Influenza Centre of Russia,
> 15/17 Professor Popov St.,
> Saint-Petersburg
> _______________________________________________
> 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 Thu Apr 19 2018 - 05:30:04 PDT
Custom Search