Hi Szymon,
Where is this underscore coming from after the SPFP? - Where it shows
"Precision Model = SPFP_"??? That should not be there.
I've never seen these problems before so I am guessing something is wrong
with your copy of AMBER in some way. I would suggest completely deleting
your tree, go back to the original tar files for the distribution. Do:
tar xvjf AmberTools13.tar.bz2
tar xvjf Amber12.tar.bz2
cd amber12
export AMBERHOME=`pwd`
./update_amber.py --update
./update_amber.py --update (yes run this twice)
./configure -cuda intel
make
make test
Note - you might want to just use the GNU compilers - the Intel compilers
are rarely tested with the GPU code and make no difference to performance.
Although I have not seen problems reported so far.
Beyond that we'll need to know more about your setup. What version of CUDA
are you using, what driver version, Linux version etc.
C1060 is also pretty old -I don't actually have any to test with anymore -
they should still work though. Note you might be better building the SPDP
precision model on such hardware though. ./configure -cuda_SPDP gnu
All the best
Ross
On 9/2/13 6:36 AM, "Szymon Murawski" <s_mur.amu.edu.pl> wrote:
>
>
>While installation of amber with cuda went flawlessly (with -cuda intel
>options) and test.cuda passed all of the implicit solvent test, all
>explicit tests generates errors:
>
>_Running Extended CUDA Explicit solvent tests._
>_ Precision Model = SPFP_
>_---------------------------------------------_
>_cd 4096wat/ && ./Run.pure_wat SPFP
>/home/users/shared/amber12/include/netcdf.mod_
>_Error: invalid argument launching kernel kNLGenerateSpatialHash_
>_ ./Run.pure_wat: Program error_
>_make[3]: [test.pmemd.cuda.pme] Error 1 (ignored)_
>_cd 4096wat_oct/ && ./Run.pure_wat_oct_NVE SPFP
>/home/users/shared/amber12/include/netcdf.mod_
>_Error: invalid argument launching kernel kNLGenerateSpatialHash_
>_ ./Run.pure_wat_oct_NVE: Program error_
>_make[3]: [test.pmemd.cuda.pme] Error 1 (ignored)_
>
>_......._
>
>_Running CUDA Explicit solvent tests._
>_ Precision Model = SPFP_
>_------------------------------------_
>_cd tip4pew/ && ./Run.tip4pew_box_nvt SPFP
>/home/users/shared/amber12/include/netcdf.mod_
>_Error: invalid argument launching kernel kNLGenerateSpatialHash_
>_ ./Run.tip4pew_box_nvt: Program error_
>_make[3]: [test.pmemd.cuda.pme.serial] Error 1 (ignored)_
>_cd tip4pew/ && ./Run.tip4pew_box_npt SPFP
>/home/users/shared/amber12/include/netcdf.mod_
>_Error: invalid argument launching kernel kCalculateCOM_
>_ ./Run.tip4pew_box_npt: Program error_
>_make[3]: [test.pmemd.cuda.pme.serial] Error 1 (ignored)_
>
>_....._
>
>_42 file comparisons passed_
>_5 file comparisons failed_
>_126 tests experienced errors_
>
>Serial and parallel installations passed all the tests, so this must be
>problem with the graphics card. My gpu is Tesla c1060, I installed all
>the latest drives and toolkits. Any idea what is causing this errors?
>
>Szymon Murawski
>
>Phd. Student
>
>Electron States of Solids Division
>
>Adam Mickiewicz University, Poznań, Poland
>
>
>_______________________________________________
>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 Mon Sep 02 2013 - 10:30:03 PDT
