On Thu, Apr 19, 2018, Kadir Ozcan wrote:
>
> After some searching it seems that Xcode had to be installed again
> using High Sierra. I was able to successfully configure CUDA with clang
> properly after reinstalling xcode. However, upon running make install I
> receive the following output leading to an error message:
>
> /Developer/NVIDIA/CUDA-9.0/bin/nvcc -gencode arch=compute_30,code=sm_30 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_52,code=sm_52 -gencode arch=compute_53,code=sm_53 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_60,code=sm_70 -Wno-deprecated-declarations -use_fast_math -O3 -ccbin clang++ -I../cusplibrary-0.6.0-rc -o cuda_cg_wrapper.o -c cuda_cg_wrapper.cu -DCUSPARSE -DPCG -DILU0 -DCSR
> nvcc fatal : Unsupported gpu architecture 'sm_70'
This is odd. What version of XCode do you have?
Try editing the config.h file and removing the
"-gencode arch=compute_60,code=sm_70" part. Does that help?
Back in January, this comment was posted:
Driver support for Mac OS 10.13.1 is added in this release. Note that
existing CUDA applications can be run on Mac OS 10.13.1 with CUDA 9.1
installed. However, CUDA 9.1 does not support building applications with
Xcode 9.1. Support for building applications with Xcode 9.1 is planned for
a future release of CUDA.
Read more at: http://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#ixzz55EgUGvDJ
Above refers to CUDA-9.1. You might have to check with NVIDIA about
what combination of XCode and nvcc you need.
(I'm replying here, simply to get things started. I'm no GPU expert,
and have never compiled pmemd.cuda on a Mac. So what I'm saying might
be way off. Nevertheless, this looks like an NVIDIA problem to me.)
....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Apr 20 2018 - 05:30:02 PDT
