From: Brian Radak (bradak_at_anl.gov)
Date: Mon Nov 14 2016 - 14:24:13 CST
This is only a partially informed answer, but I believe the answer is
yes. Specifically I'm pretty sure the targeted MD calculations require a
CPU only computation at each cycle - this will be considerably slower
than other MD methods.
In general, anything that requires a system-wide calculation (such as
the energy or some collective variable) will drastically reduce
performance. You will notice that this also holds for high frequency I/O
(e.g. setting outputEnergies to a small value).
On 11/14/2016 04:11 AM, sonia ziada wrote:
> Dear NAMD users,
> I am running NPT brute force molecular dynamics simulation with PBC on
> a system of 100 000 atoms using 1 nvidia K80 gpu card.
> The best performance (16.7ns/day) is obtained for a number of cpu
> equal to 8 (+p8) and the GPU-utility is around 60%.
> If I run a targeted molecular dynamics simulation, with the same
> system and the same condition, i get a performance of 2.8ns/day
> (against 1.81ns/day using only 8 cpu without GPU card) and th
> GPU-utility is around 12%.
> Why the improvement in performance is so bad when switching from CPU
> to GPU for the targeted molecular dynamics (TMD) simulation ? Is it
> normal that the GPU utility is only 12% ? Are there features badly
> suuported on GPU ? (like here for the TMD)
-- Brian Radak Postdoctoral Appointee Leadership Computing Facility Argonne National Laboratory 9700 South Cass Avenue, Bldg. 240 Argonne, IL 60439-4854 (630) 252-8643 brian.radak_at_anl.gov
This archive was generated by hypermail 2.1.6 : Tue Dec 27 2016 - 23:22:36 CST