Hi Vitor,
Apologies for the long delay here. I finally got around to testing this
properly. There is indeed a bug in AMBER 11 GPU with regards to the use of
organic solvents. This is now fixed in our development tree and I hope to
put a patch together, with other fixes and performance enhancements bundled
together in a few weeks.
All the best
Ross
> -----Original Message-----
> From: vitor.felix.ua.pt [mailto:vitor.felix.ua.pt]
> Sent: Monday, February 28, 2011 1:05 AM
> To: amber.ambermd.org
> Subject: [AMBER] NPT simulation crash using a GPU CARD for a simulation
> carried out in a organic explicit solvent
>
> Dear Amber community
>
> I am interested to run a MD simulation in a NPT ensemble using a NVIDIA
> GPU C2050. The solvent is a mixture composed of methanol and
> chloroform
> and the periodic system contains only 7858 atoms. The simulation runs
> well in both NVT and NVE ensembles.When the NPT is used, the simulation
> stops in the begging with message given below. Furthermore, the same
> problem occurs with a pure solvent such as chloroform. In contrast ,
> when the mixture of solvent is composed of water and an organic
> solvent (the system has approximately the same size in number of
> atoms) the NPT simulation runs without any problems as happened with
> much larger size simulations carried in TIP3P water model. The problem
> seems to me related with the type of explicit solvent used. The system
> is quite small and I donīt believe that the problem is related with
> amount of GPU memory required by a NPT simulation.
> Searching on the internet I found the same message type for other GPU
> cards, but not for Tesla C2050.
>
> Any help would be appreciated about this issue.
>
> Thank you very much in advance
> Vitor Felix
>
> #######################################################
> Error: unspecified launch failure launching kernel kNLSkinTest
> cudaFree GpuBuffer::Deallocate failed unspecified launch failure
>
> Begin of the NPT simulation
> ---------------------------------------------------
> APPROXIMATING switch and d/dx switch using CUBIC SPLINE INTERPOLATION
> using 5000.0 points per unit in tabled values
> TESTING RELATIVE ERROR over r ranging from 0.0 to cutoff
> | CHECK switch(x): max rel err = 0.2738E-14 at 2.422500
> | CHECK d/dx switch(x): max rel err = 0.7967E-11 at 2.716640
> ---------------------------------------------------
> |---------------------------------------------------
> | APPROXIMATING direct energy using CUBIC SPLINE INTERPOLATION
> | with 50.0 points per unit in tabled values
> | Relative Error Limit not exceeded for r .gt. 2.27
> | APPROXIMATING direct force using CUBIC SPLINE INTERPOLATION
> | with 50.0 points per unit in tabled values
> | Relative Error Limit not exceeded for r .gt. 2.78
> #######################################################################
> #############
>
> --
> *************************************
> Vitor Felix
> Department of Chemistry
> University of Aveiro
> 3810-193 Aveiro, Portugal
> Tel: 351 234 370 200 ext: 22102
> http://molecular-modeling.dq.ua.pt
>
>
> _______________________________________________
> 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 Wed May 04 2011 - 09:30:03 PDT
