From: Giacomo Fiorin (giacomo.fiorin_at_gmail.com)
Date: Wed Oct 14 2020 - 09:15:30 CDT
Using Colvars instead of the much older TMD implementation makes sense in
general, but I'm not 100% sure here.
Enabling Colvars in NAMD 3 on even a few atoms would probably damage
performance, because of the latency associated with the transmission of
data between GPU and CPU. On the positive side, it should be easy to test
if (1) it works and (2) performance is still acceptable on a short
benchmark.
FYI it looks like the OpenMM folks have struggled with a similar problem
and eventually implemented several collective variables directly as
internal objects, instead of using PLUMED as they had done previously.
Giacomo
On Wed, Oct 14, 2020 at 9:50 AM Jérôme Hénin <jerome.henin_at_ibpc.fr> wrote:
> Hi Harper,
>
> I don't know if the NAMD implementation of TMD is supposed to be supported
> in NAMD 3 and beyond, but you can always perform TMD using the Collective
> Variables (Colvars) Module, with a moving restraint on a RMSD coordinate.
>
> Best,
> Jerome
>
> ----- On 13 Oct 20, at 18:18, Smith, Harper E.
> smith.12510_at_buckeyemail.osu.edu wrote:
>
> > Hi mailing list,
> >
> > I want to use NAMD 3.0 alpha 6 with CUDASOAintegrate on to run a
> targeted MD
> > simulation. Although I see TMD output in NAMD 2.12, I do not see it after
> > switching.
> >
> > I get several lines that it is active:
> >
> > Info: TMD ACTIVE BETWEEN STEPS 0 and 1250000
> > Info: TMD K 200
> > Info: TMD FILE trimer-post-targeting.tmd
> > Info: TMD OUTPUT FREQUENCY 1000
> > Info: TMD TARGET RMSD AT FIRST STEP -1
> > Info: TMD TARGET RMSD AT FINAL STEP 0
> >
> > .but the relevant output is missing. Is TMD supported in NAMD 3.0?
> >
> > Best,
> > Harper Smith
>
>
This archive was generated by hypermail 2.1.6 : Thu Dec 31 2020 - 23:17:14 CST