Dear Charles
The machines we have jointly designed with Exxact should be suitable:
http://ambermd.org/gpus/recommended_hardware.htm#exxact_machines
Specifically one can customize the CPU and so you could for example
request 8 core, 3.0GHz CPUs instead of cheaper 6 or 4 core variants. Even
with high spec CPUs a single 4 x GTX680 node will come in under $8K or so.
Less if you request lower specced CPUs. The issue is likely to be more how
well the various codes run on GTX680 cards which have very limited double
precision compared with the Kepler K20 cards. The later will add
significantly to the cost. You might want to check with the various code
authors about the suitability of GTX680 cards for their applications. My
guess would be that Gromacs should be fine with GTX680s, last time I
checked they were using OpenMM which is single precision. NAMD might
require K20s since they may be using double precision on the GPUs. I have
not tried this explicitly though so am only speculating.
My advice would be to sign up for the free GPU test drive with Exxact
here: http://exxactcorp.com/testdrive/AMBER/
They have both 4xGTX680 and 4xK20 machines available, with 8 core E5-2660
CPUs. That way you can try running both AMBER and Gromacs on these two
machines and thus decide for yourself if you will need the K20 cards or
whether GTX680s will suffice.
I hope that helps.
All the best
Ross
On 2/19/13 7:12 PM, "Charles Do" <cdonssim.gmail.com> wrote:
>Hi
>
>Is this Kepler GTX680 MD Workstation generally good for any MD simulation
>softwares such as GROMACS ? My first guess is no, because GROMACS highly
>depends on CPUs than GPUs. Is there a cost-effective small size machine
>good for general simulation works (both AMBER and GROMACS ) under $20,000
>?
>
>Thank you,
>
>C. Do
>_______________________________________________
>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 Tue Feb 19 2013 - 20:00:02 PST
