Downgrade to an older cuda.
I know there is a pretty serious bug in Nvidia's cufft library in CUDA 7 anyway (specifically regarding asynchronous streams). I don't know if pmemd.cuda uses those, but in any case I'd suggest downgrading for now.
I use CUDA 6.5, personally.
HTH,
Jason
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
> On Apr 10, 2015, at 4:38 PM, Emilio Angelina <emilioluisangelina.hotmail.com> wrote:
>
> Dear AMBER users,
>
> I am trying to install Amber12 with cuda. My Graphic card is a geforce GTX970.
> I have installed CUDA toolkit 7.0.
> When I try to install Amber I get the following error
>
> Error: Unsupported CUDA version 7.0 detected.
>
>
>>> AMBER requires CUDA version == 4.2 .or. 5.0 .or. 5.5
>
>
>>> Configure failed due to the errors above!
>
>
> Following posts in this Mailing List I commented out the check in $AMBERHOME/AmberTools/configure2 but I had no luck. Configure script ends correctly but after running make install I get:
>
> nvcc fatal : Unsupported gpu architecture 'compute_13'
> make[4]: *** [kForcesUpdate.o] Error 1
> make[4]: Leaving directory `/home/angelina/amber/src/pmemd/src/cuda'
> make[3]: *** [cuda/cuda.a] Error 2
> make[3]: Leaving directory `/home/angelina/amber/src/pmemd/src'
> make[2]: *** [cuda] Error 2
> make[2]: Leaving directory `/home/angelina/amber/src/pmemd'
> make[1]: *** [cuda] Error 2
> make[1]: Leaving directory `/home/angelina/amber/src'
> make: *** [install] Error 2
>
> How I can circumvent this problem? There is a way of solving it without having to upgrade to Amber14?
>
>
> Thanks in advance
> Emilio
>
>
> _______________________________________________
> 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 Fri Apr 10 2015 - 15:00:02 PDT