Re: [AMBER-Developers] pmemd.cuda errors (run-time, not compilation)

From: Jason Swails <jason.swails.gmail.com>
Date: Wed, 10 Mar 2010 12:39:51 -0500

On Wed, Mar 10, 2010 at 12:07 PM, Ross Walker <ross.rosswalker.co.uk> wrote:
> Hi Jason,
>
> Just to double check...
>
> 1) Are you using the driver AND cuda version that go together?

gahh, didn't think about this. I don't have permissions to change the
driver (it's the Tesla machine on the local HPC here). They used to
have one node running 3.0, but then changed it back to 2.3 for some
reason...

Thanks,
Jason

>
> E.g. if you installed the Release candidate driver you should use the
> release candidate compiler. If you are using the beta compiler from the
> website you should use the beta compiler from the same page.
>
> 2) Did you reboot after changing the graphics driver?
>
> 3) Are you certain you are using the same binary model in all cases? E.g. an
> x86_64 OS + x86_64 version of gfortran and gcc + x86_64 versions of the cuda
> driver and compiler?
>
> 4) Are you using the default precision model?
>
> Specifying -cuda_DPDP or -cuda_SPSP could have 'undefined' results right
> now.
>
> All the best
> Ross
>
>> -----Original Message-----
>> From: amber-developers-bounces.ambermd.org [mailto:amber-developers-
>> bounces.ambermd.org] On Behalf Of Jason Swails
>> Sent: Wednesday, March 10, 2010 8:55 AM
>> To: AMBER Developers Mailing List
>> Subject: [AMBER-Developers] pmemd.cuda errors (run-time, not
>> compilation)
>>
>> Hello,
>>
>> I'm having some trouble with pmemd.cuda.  There is a bit of confusion
>> with the different cuda versions, but I'm getting the same error with
>> cuda 3.0 on Ross's website (changing #if 1 to #if 0 to get rid of the
>> hack) and the beta version on the website where I keep the hack.
>>
>> [swails.tesla1 test]$ make test.pmemd.cuda
>> cd cuda && make -k test.pmemd.cuda
>> make[1]: Entering directory
>> `/scratch/ufhpc/swails/newamber/03.09.10/amber11/test/cuda'
>> export TESTsander='../../../exe/pmemd.cuda'; cd gb_ala3/ &&
>> ./Run.igb1_ntc1_min
>> cudaMalloc GpuBuffer::Allocate failed unspecified driver error
>>   ./Run.igb1_ntc1_min:  Program error
>> make[1]: *** [test.pmemd.cuda.gb] Error 1
>> export TESTsander='../../../exe/pmemd.cuda'; cd 4096wat/ &&
>> ./Run.pure_wat
>> cudaMemcpyToSymbol: SetSim copy to cSim failed unspecified driver error
>>   ./Run.pure_wat:  Program error
>> make[1]: *** [test.pmemd.cuda.pme] Error 1
>> make[1]: Target `test.pmemd.cuda' not remade because of errors.
>> make[1]: Leaving directory
>> `/scratch/ufhpc/swails/newamber/03.09.10/amber11/test/cuda'
>> make: *** [test.pmemd.cuda] Error 2
>>
>> nvcc does point to the one in the 3.0 release, and $CUDA_HOME/lib64
>> appears in my LD_LIBRARY_PATH before the 2.3 version on the system.  I
>> also get the same error regardless of how I build it (SPDP, SPSP, DPDP
>> in both the pmemd Makefile and amber's Makefile).
>>
>> Any ideas?
>>
>> Thanks!
>> Jason
>>
>> --
>> ---------------------------------------
>> Jason M. Swails
>> Quantum Theory Project,
>> University of Florida
>> Ph.D. Graduate Student
>> 352-392-4032
>>
>> _______________________________________________
>> AMBER-Developers mailing list
>> AMBER-Developers.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber-developers
>
>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
>



-- 
---------------------------------------
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Graduate Student
352-392-4032
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Wed Mar 10 2010 - 10:00:04 PST
Custom Search