Re: [AMBER] Configure CUDA issues

From: David A Case <>
Date: Thu, 3 May 2018 13:35:32 -0400

On Thu, May 03, 2018, Kadir Ozcan wrote:
> I recently upgraded to Amber18. I installed Amber18 as well as
> AmberTools18 successfully. I want to configure CUDA so I can GPU
> accelerate pmemd. I have a MacOS 10.13.4 (17e199). I am using an
> external GPU which is GeForce GTX 980 Ti and I have installed CUDA
> Toolkit 9.0. Am I using the correct CUDA? I ran the following code:
> ./configure -cuda clang
> make install
> My first error was that my compiler version was not supported (90100) so
> I switched back to Xcode 8.2 which resolved that issue. My next error
> was "nvcc fatal : Unsupported gpu architecture ‘sm_70’". I removed the
> portion "-gencode arch=compute_70,code=sm_70” from “config.h” in the
> Amber18 directory as suggested by David A. Case.

>From the lack of other responses, it seems likely that no one following
this list has used pmemd.cuda on a Mac.

> I then try “make install” again and receive the following error:
> nvcc fatal : Unsupported gpu architecture 'sm_70'
> make[2]: *** [build_cpptraj] Error 2

Your error is in building the cuda version of cpptraj, and you would
have to also modify the config.h in $AMBERHOME/AmberTools/src/cpptraj.

However, since you want pmemd.cuda, cd to $AMBERHOME/src/pmemd, and type

   make cuda_serial

That will pick up your modified config.h info, and might get one step
closer to resolution.


AMBER mailing list
Received on Thu May 03 2018 - 11:00:03 PDT
Custom Search