AMBER Archive (2008)

Subject: RE: Fw: RE: AMBER: MKL libraries/Amber10

From: Francesco Pietra (chiendarret_at_yahoo.com)
Date: Fri May 30 2008 - 04:40:41 CDT

Hi Ross and Cenk:

If its is MKL that makes the problem (on different hardware, by Cenk and me), why not compiling Amber10 with ifort + ACML or GOTO libraries? Is any indication how to set the congif file?

In previous mail Ross said that MKL will speed up QMMM by a large (not minor) margin. As far as I know ACML are equivalent BLAS, and perhaps GOTO is even better than MKL or ACML.

I am not pressed to use QMMM, though I plan the future.

Thanks

francesco

--- On Thu, 5/29/08, Ross Walker <ross_at_rosswalker.co.uk> wrote:

> From: Ross Walker <ross_at_rosswalker.co.uk>
> Subject: RE: Fw: RE: AMBER: MKL libraries/Amber10
> To: amber_at_scripps.edu
> Date: Thursday, May 29, 2008, 9:50 PM
> Hi Cenk,
>
> I have found on bug in parallel which I will put together a
> bugfix for
> although I don't think it is what is causing the
> segfaults. I think the
> segfaults are actually coming from deep inside MKL itself
> so it may actually
> be a problem with a specific version of MKL - perhaps on
> specific hardware,
> but I can't work out why it would only occur in
> parallel. I will explore
> some more tomorrow and let you know. For the time being I
> think you will be
> fine if you build a second version of amber without MKL and
> use that for
> QMMM. It will be a little slower but it should pass all the
> test cases.
>
> All the best
> Ross
>
> > -----Original Message-----
> > From: owner-amber_at_scripps.edu
> [mailto:owner-amber_at_scripps.edu] On Behalf
> > Of Cenk (Jenk) Andac
> > Sent: Thursday, May 29, 2008 5:25 PM
> > To: amber_at_scripps.edu
> > Subject: RE: Fw: RE: AMBER: MKL libraries/Amber10
> >
> > Hi Ross,
> >
> > I was wondering if you have had some time to think
> about what might be
> > actually happening in the QMMM test failures I have
> had with the
> > sander.MPI module of AMBER10. Any suggestions would be
> welcome.
> >
> > My best regards,
> >
> > jenk
> >
> >
> > --- On Thu, 5/22/08, Ross Walker
> <ross_at_rosswalker.co.uk> wrote:
> >
> > > From: Ross Walker <ross_at_rosswalker.co.uk>
> > > Subject: RE: Fw: RE: AMBER: MKL libraries/Amber10
> > > To: amber_at_scripps.edu
> > > Date: Thursday, May 22, 2008, 5:40 PM
> > > Hi Cenk,
> > >
> > > Can you possibly try something for me.
> > >
> > > Can you run the parallel test cases on both one
> of the
> > > nodes in parallel and
> > > on the login node in parallel (if it will let
> you).
> > > Specifically I want to
> > > find out if the problem is processor specific. I
> only have
> > > access to Intel
> > > chips at the moment and everything works fine. It
> is
> > > possible that MKL's
> > > dspev crashes on AMD chips (I wouldn't put it
> past
> > > Intel ;-)).
> > >
> > > Thus I am guessing that this may be a problem
> specific to
> > > AMD chips. I'll
> > > try and get on Ranger to try this myself but it
> has been
> > > down for the last
> > > few days.
> > >
> > > All the best
> > > Ross
> > >
> > >
> > >
> -----------------------------------------------------------------------
> > > The AMBER Mail Reflector
> > > To post, send mail to amber_at_scripps.edu
> > > To unsubscribe, send "unsubscribe
> amber" (in the
> > > *body* of the email)
> > > to majordomo_at_scripps.edu
> >
> >
> >
> >
> -----------------------------------------------------------------------
> > The AMBER Mail Reflector
> > To post, send mail to amber_at_scripps.edu
> > To unsubscribe, send "unsubscribe amber" (in
> the *body* of the email)
> > to majordomo_at_scripps.edu
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" (in the
> *body* of the email)
> to majordomo_at_scripps.edu

      
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" (in the *body* of the email)
      to majordomo_at_scripps.edu