Hi Melinda,
Is this the only test case that fails or do other GB test cases fail? This
is indeed too big a difference and is why we have the test cases so people
spot these otherwise 'silent' errors. Note the differences are all in the
electrostatics - 1-4 EEL, EGB and EELEC. This suggests a possible compiler
bug - probably a misvectorization of a loop - inside the EGB routine.
What compiler are you using and what processor is it?
I currently use ifort v10.1.018 on Intel E5462 dual quad core system and all
the test cases pass. This is using MKL (v10.0.5.025), I have not tried it
without MKL.
I would suggest trying a new compiler version if you can (letting us know
what one is causing the problem). We could also try working around the
problem if we can locate where things are going wrong but I am loathed to do
this if simply upgrading the compiler fixes it.
All the best
Ross
> -----Original Message-----
> From: owner-amber.scripps.edu [mailto:owner-amber.scripps.edu] On Behalf
> Of Melinda Layten
> Sent: Monday, October 20, 2008 1:23 PM
> To: amber.scripps.edu; jmongan.mccammon.ucsd.edu
> Subject: AMBER: cnstph test case failure or expected divergence?
>
> I'm in the process of installing Amber on two machines, an Intel core
> 2 quad core server running Debian and a Sun Solaris installation.
>
> On the sun, the cnstph test case is running as expected and producing
> identical results to the standard.
>
> On our Intel server, even the first step is producing drastically
> different energies, especially EGB and by step 20 the states have
> changed significantly from the standard. Parallel and serial
> implementations are producing identical results, this just seems too
> different to be a rounding change. Is this just a randomization
> change or should we be looking deeper.
>
> Thanks
>
> Melinda Layten
>
> cnstph energy outputs:
> Standard/Sun:
> NSTEP = 0 TIME(PS) = 0.000 TEMP(K) = 355.72 PRESS =
> 0.0
> Etot = -4105.8987 EKtot = 1762.6242 EPtot = -
> 5868.5229
> BOND = 64.4710 ANGLE = 230.5518 DIHED =
> 830.6145
> 1-4 NB = 415.5580 1-4 EEL = 3765.5754 VDWAALS = -
> 1096.5582
> EELEC = -8750.9534 EGB = -1327.7821 RESTRAINT =
> 0.0000
> -------------------------------------------------------------------------
> -----
>
>
> NSTEP = 10 TIME(PS) = 0.020 TEMP(K) = 151.38 PRESS =
> 0.0
> Etot = -4409.5041 EKtot = 750.0758 EPtot = -
> 5159.5799
> BOND = 198.3337 ANGLE = 688.7830 DIHED =
> 924.9432
> 1-4 NB = 451.5455 1-4 EEL = 3774.6634 VDWAALS = -
> 1067.8172
> EELEC = -8649.0618 EGB = -1480.9696 RESTRAINT =
> 0.0000
> -------------------------------------------------------------------------
> -----
>
> Intel:
>
> NSTEP = 0 TIME(PS) = 0.000 TEMP(K) = 355.72 PRESS =
> 0.0
> Etot = -3871.7448 EKtot = 1762.6242 EPtot = -
> 5634.3690
> BOND = 64.4710 ANGLE = 230.5518 DIHED =
> 830.6145
> 1-4 NB = 415.5580 1-4 EEL = 3623.6672 VDWAALS = -
> 1096.5582
> EELEC = -6604.1032 EGB = -3098.5702 RESTRAINT =
> 0.0000
> -------------------------------------------------------------------------
> -----
>
>
> NSTEP = 10 TIME(PS) = 0.020 TEMP(K) = 153.51 PRESS =
> 0.0
> Etot = -4158.6813 EKtot = 760.6689 EPtot = -
> 4919.3502
> BOND = 202.1132 ANGLE = 691.0621 DIHED =
> 938.7263
> 1-4 NB = 452.6111 1-4 EEL = 3586.3736 VDWAALS = -
> 1070.0638
> EELEC = -6567.6242 EGB = -3152.5485 RESTRAINT =
> 0.0000
> -------------------------------------------------------------------------
> -----
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber.scripps.edu
> To unsubscribe, send "unsubscribe amber" (in the *body* of the email)
> to majordomo.scripps.edu
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" (in the *body* of the email)
to majordomo.scripps.edu
Received on Wed Oct 22 2008 - 05:08:23 PDT
