On Thu, Mar 18, 2010 at 3:48 PM, Ross Walker <ross.rosswalker.co.uk> wrote:
> Hi Jason,
>
>> I have a quick question about pmemd.cuda and running it on various
>> nVidia cards. I'm getting the message "No double-precision capable
>> gpu located, exiting." even though I compiled with SPSP options (which
>> makes sense since that message is not located within any kind of ifdef
>> statement). Does this mean that card is unsuitable for pmemd.cuda (in
>> other words, double-precision is required even though I specified
>> single-single?)
>
> The SPSP option is for DEBUGGING ONLY. Do NOT use it for production
> calculations. Also note that it still does shake in double precision since
> you need 10^-6 or better tolerances. I will be writing documentation to this
> effect along with manuscripts showing the differences between the various
> precision models when I get a chance to actually start removing things from
> my TODO list faster than they accumulate. For the moment the use of single
> precision for the entire calculation has not been fully characterized and
> indeed blows up for systems where the box size gets large - note why several
> benchmarks in the literature that use just single precision for 500K+ atom
> systems only show timings for 50 steps. ;-)
>
> Short answer is only cards with SM version 1.3 or later are supported. That
> is GT-2XX series and C10XX series or Fermi.
The card I'm working on is a GTS-250... Smells like a supported card,
but I'm not sure.
>
> Last time I checked the SPSP and DPDP options are not documented in the
> configure script.
>
> -cuda Builds the NVIDIA GPU accelerated version of pmemd
> (pmemd.cuda).
> with default SPDP hybrid precision.
> (Note: this option is exclusive of mpi and only works with
> gnu
> and intel at present. CUDA_HOME must point to your cuda
> build
> tools installation path.).
>
> Where did you find these options? It is important that users do not
> arbitrarily start messing with such things so if it is in some obvious
> documentation somewhere I'd like to remove it.
>
> All the best
> Ross
>
> /\
> \/
> |\oss Walker
>
> | Assistant Research Professor |
> | San Diego Supercomputer Center |
> | Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
> | http://www.rosswalker.co.uk | http://www.wmd-lab.org/ |
>
> Note: Electronic Mail is not secure, has no guarantee of delivery, may not
> be read every day, and should not be used for urgent or sensitive issues.
>
>
>
>
>
>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
>
--
---------------------------------------
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Graduate Student
352-392-4032
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Thu Mar 18 2010 - 13:00:04 PDT
