Re: [AMBER] run_cmake error for installing the GPU version of Amber20

From: David A Case <david.case.rutgers.edu>
Date: Sun, 5 Dec 2021 08:35:24 -0500

On Sun, Dec 05, 2021, jinfeng liu wrote:
>
>I am trying to install Amber20 on my centos7 server. The serial version has
>been installed successfully. But when turn to the GPU version, I came
>across an wired error message like the following:
>
>Could NOT find CUDA (missing: CUDA_INCLUDE_DIRS CUDA_CUDART_LIBRARY) (found
>version "11.1")
>CMake Error at cmake/CudaConfig.cmake:15 (message):
> You turned on CUDA, but it was not found. Please set the
> CUDA_TOOLKIT_ROOT_DIR option to your CUDA install directory.

You could try two things:

1. Do what the error message suggests: add the following line to your
run_cmake script:

    -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-11.3

2. Running a Google search on "cmake could not find cuda" leads to the
following top hit:

    https://stackoverflow.com/questions/19980412/how-to-let-cmake-find-cuda

This suggests several alternatives to solution 1:

     a. set the CUDA_BIN_PATH environement variable (not CUDA_HOME -- that's
        our fault)
     b. make sure that nvcc is in your PATH (not sure why this didn't work
        for you, but double-check that "which nvcc" gives what you want)
     c. make a symlink from /usr/local/cuda-11.3 to /usr/local/cuda; check
        that there is *not* symlink between /usr/local/cuda and any other
        version of the CUDA toolkit

...good luck...dac


_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sun Dec 05 2021 - 06:00:02 PST
Custom Search