Could NOT find NVML (missing: NVML_LIBRARIES) (found version "10")

[bpinaya]

Under the lastest Jetpack version 4.2.1 for embedded devices like Xavier or TX2 I'm seeing this issue when trying to cmake:

cmake .. -DNVML_RETRIEVE_HEADER_ONLINE=True

I get the error:

-- The C compiler identification is GNU 7.4.0
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
CMake Error at /usr/local/share/cmake-3.15/Modules/FindPackageHandleStandardArgs.cmake:137 (message):
  Could NOT find NVML (missing: NVML_LIBRARIES) (found version "10")
Call Stack (most recent call first):
  /usr/local/share/cmake-3.15/Modules/FindPackageHandleStandardArgs.cmake:378 (_FPHSA_FAILURE_MESSAGE)
  cmake/modules/FindNVML.cmake:52 (FIND_PACKAGE_HANDLE_STANDARD_ARGS)
  CMakeLists.txt:31 (find_package)


-- Configuring incomplete, errors occurred!
See also "/home/nvidia/Documents/nvtop/build/CMakeFiles/CMakeOutput.log".

It seems it finds NVML and doesn't find it in the download part where you look for:

https://raw.githubusercontent.com/NVIDIA/nvidia-settings/master/src/nvml.h

Any insight is appreciated!

[Syllo]

CMake complains because it cannot find the library (libnvml.so file), not the header.
Is the Nvidia proprietary driver installed?

[rmrao]

I'm getting the same issue on only one machine (mine says found version "9"). The driver is definitely installed and I can run nvidia-smi. I can't figure out what the difference is - I've installed it on two other machines with no problems. It looks like the file is located at /usr/lib/x86_64-linux-gnu/libnvidia-ml.so on all my machines.

[lamhoangtung]

I'm getting the same issue on only one machine (mine says found version "9"). The driver is definitely installed and I can run nvidia-smi. I can't figure out what the difference is - I've installed it on two other machines with no problems. It looks like the file is located at /usr/lib/x86_64-linux-gnu/libnvidia-ml.so on all my machines.

If you have a conda enable. Try to ‘conda deactivate’ first before you make nvtop (even if conda at base environment)

[rmrao]

Awesome this worked. I guess my compilers were set to the conda compilers on this one machine. Thanks for the suggestion!

[Syllo]

Thank you @lamhoangtung for providing a solution.

I added a comment about conda in the README, hopefully this will help.

[lamhoangtung]

Thank you @lamhoangtung for providing a solution.

I added a comment about conda in the README, hopefully this will help.

Thanks ^^. Looks like we can close the issue for now

[sytelus]

Here's the full script that takes care of deactivating conda and the reactivating it after install. It would be great to add this in git as something like install_ubuntu18.sh so people don't have to keep typing same stuff over and over.

sudo apt -y install cmake libncurses5-dev libncursesw5-dev git

# clone if not already exist
[ ! -d 'nvtop' ] && git clone https://github.com/Syllo/nvtop.git
mkdir -p nvtop/build && cd nvtop/build

# if conda environment then first deactivate it
env="${CONDA_DEFAULT_ENV}"
if [[ ! -z "${env}" ]]; then
    # https://github.com/conda/conda/issues/7980#issuecomment-524154596
    eval "$(conda shell.bash hook)"
    conda deactivate
fi
cmake ..

# If it errors with "Could NOT find NVML (missing: NVML_INCLUDE_DIRS)"
# try the following command instead, otherwise skip to the build with make.
cmake .. -DNVML_RETRIEVE_HEADER_ONLINE=True

make
sudo make install # You may need sufficient permission for that (root)

# reactivate conda environment
if [[ ! -z "${env}" ]]; then
    conda activate "${env}"
fi

[andviane]

For me did not: /bin/sh: 1: conda: not found . The library itself should be installed, because I run sudo apt-get install cuda-nvml-dev-10-1. It is also only on one machine. It can be observed when building on the docker image nvidia/cuda:10.1-cudnn7-devel-ubuntu18.04 that should have all CUDA install (and can run apt-get install cuda-nvml-dev-10-1 without errors).

[pharpe]

I'm also getting
CMake Error at /usr/share/cmake-3.10/Modules/FindPackageHandleStandardArgs.cmake:137 (message): Could NOT find NVML (missing: NVML_LIBRARIES) (found version "10")

I have tried cmake .. -DNVML_RETRIEVE_HEADER_ONLINE=True and get the same error. I don't have conda installed. I also tried running the script that @sytelus posted and get the same error.

Here is the log;

CMakeOutput.log

[snafu123]

i see this too

[eriksneff]

@pharpe I'm seeing the same thing - Could NOT find NVML (missing: NVML_LIBRARIES) (found version "10"). I'm on a stock install on the jetson xavier mx dev kit board. I don't have conda installed. Just looking for a quick way to montior gpu utilization as I test transcoding performance. Frustrating.

[Syllo]

The only thing I can think of at this point is adding more hints for CMake.

Can you please try with the version on the branch cmake_nvml_notfound.

If this does not work, can you please provide the path where the library is located and I will add it to the hint list.

[pharpe]

Removed the old git clone

sudo rm -rIf nvtop

cloned the cmake_nvml_notfound branch

git clone -b cmake_nvml_notfound --single-branch https://github.com/Syllo/nvtop.git

Created the build directory and cd to it

mkdir -p nvtop/build && cd nvtop/build

Ran cmake

user@server:~/nvtop/build$ cmake ..
-- The C compiler identification is GNU 7.5.0
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Setting build type to 'Release' as none was specified.
-- Could NOT find NVML (missing: NVML_INCLUDE_DIRS NVML_LIBRARIES)
CMake Error at /usr/share/cmake-3.10/Modules/FindPackageHandleStandardArgs.cmake:137 (message):
  Could NOT find NVML (missing: NVML_INCLUDE_DIRS NVML_LIBRARIES)
Call Stack (most recent call first):
  /usr/share/cmake-3.10/Modules/FindPackageHandleStandardArgs.cmake:378 (_FPHSA_FAILURE_MESSAGE)
  cmake/modules/FindNVML.cmake:52 (FIND_PACKAGE_HANDLE_STANDARD_ARGS)
  CMakeLists.txt:33 (find_package)

I tried

cmake .. -DNVML_RETRIEVE_HEADER_ONLINE=True

user@server:~/nvtop/build$ cmake .. -DNVML_RETRIEVE_HEADER_ONLINE=True
-- Could NOT find NVML (missing: NVML_INCLUDE_DIRS NVML_LIBRARIES)
CMake Error at /usr/share/cmake-3.10/Modules/FindPackageHandleStandardArgs.cmake:137 (message):
  Could NOT find NVML (missing: NVML_INCLUDE_DIRS NVML_LIBRARIES)
Call Stack (most recent call first):
  /usr/share/cmake-3.10/Modules/FindPackageHandleStandardArgs.cmake:378 (_FPHSA_FAILURE_MESSAGE)
  cmake/modules/FindNVML.cmake:52 (FIND_PACKAGE_HANDLE_STANDARD_ARGS)
  CMakeLists.txt:33 (find_package)

-- Configuring incomplete, errors occurred!
See also "/home/tom/nvtop/build/CMakeFiles/CMakeOutput.log".

[Syllo]

Thanks @pharpe, is your libnvidia-ml.so library installed in one of the following directories? /usr/lib64/nvidia*/ /usr/lib/nvidia*/ /usr/lib/*linux-gnu

[benroeder]

I am having the same issue, libnvidia-ml.so is installed in
/usr/local/cuda-10.2/targets/aarch64-linux/lib/stubs/libnvidia-ml.so

[Mistobaan]

fixed with my patch https://github.com/Mistobaan/nvtop#master and adding. This washappening building from an nvidia TensorRT container

cmake .. \
    -DNVML_INCLUDE_DIRS=/usr/local/cuda-10.1/targets/x86_64-linux/include/ \
    -DNVML_LIBRARIES=/usr/local/cuda-10.1/targets/x86_64-linux/lib/stubs/libnvidia-ml.so

[Syllo]

Alright, I added some hints to help CMake find the library in the patch 5fb0ab5.
This should help when cuda is installed under the /usr/local/cuda* hierarchy.

[Syllo]

Close as we don't rely on the nvml header anymore.