Re: [AMBER] Major GPU Update Released

From: Scott Le Grand <varelse2005.gmail.com>
Date: Sun, 21 Aug 2011 13:34:35 -0700

Try my patch. And if that doesn't work, divide by 4 instead of 2...

Plenty of headroom here to experiment... Though you guys really need to
spend a little more money on your GPUs IMO... This is just getting silly if
you ask me (not that you did)...


On Sun, Aug 21, 2011 at 1:27 PM, Dmitry Mukha <dvmukha.gmail.com> wrote:

> 1GB on board
>
> 51092 atoms
> 46110 atoms belong to water (15370 molecules)
> 4976 atoms of protein (2503 heavy atoms)
>
> BEFORE APPLYING THE PATCH pmemd.cuda had performance 3.84 ns/day (285) and
> slightly more than 12 ns for 580.
> NOW launch failed (segfault)
>
> Langevin thermostat (2 ps-1) + pressure control (relaxation time 2 ps).
> Switching to Berendsen thermostat (and removing gamma_ln) allows to execute
> pmemd.cuda successfully. Again, this system plays a role of test, because
> it
> ran well before.
>
> Previous version printed error message about lack of memory in cases when
> system was too large, so the error I have faced is definitely new. 'Line 9'
> in the error message corresponds to the code of executable file, not script
> (it is occurred in the line 18 of bash script).
>
> 2011/8/21 Scott Le Grand <varelse2005.gmail.com>
>
> > Questions:
> > GPU? Assuming 285 with 2 GB? (don't actually have one anymore as they
> all
> > died of old age)
> > Number of atoms?
> >
> > > >
> > > > In my case GPU memory isn't loaded very much. nvidia-smi reports
> about
> > > 16%
> > > is utilized with ntt=1, that means it will be enough space on board.
> > tests
> > > were passed (if memory amount was sufficient), but simulations that
> went
> > > fine previously now do not (re-)launch because of that segfault...
> > > Amber11 / at15 / patches till 17th...
> > >
> >
> --
> Sincerely,
> Dmitry Mukha
> Institute of Bioorganic Chemistry, NAS, Minsk, Belarus
> _______________________________________________
> 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 Sun Aug 21 2011 - 14:00:02 PDT
Custom Search