Configuring GROMACS (patched with PLUMED) with GPU support in Linux

GROMACS + PLUMED + GPU + LINUX

By SAJEEWA PEMASINGHE

System description

CPU - Intel(R) Core(TM) i7-7500 @2.70 GHz
GPU chip - NVIDIA GeForce 930MX
OS - Ubuntu 16.04 LTS 64 bit

Setting up the NVIDIA driver and CUDA

#Download NVIDIA-Linux-x86_64-384.111.run from  http://www.nvidia.com/Download/driverResults.aspx/128737/en-us
#Download cuda_8.0.61_375.26_linux.run  from https://developer.nvidia.com/cuda-80-ga2-download-archive
#Remove previous versions (if applicable to you)

sudo apt-get --purge remove nvidia-*
sudo nvidia-uninstall

#press ctrl+alt+F2 to go to a tty session and login with your username and password

sudo service lightdm stop

#Blacklist nouveau drivers 
# Create a file named blacklist-nouveau.conf inside /etc/modprobe.d with the following content

blacklist nouveau
options nouveau modeset=0

#Regenerate the kernel initramfs

sudo update-initramfs -u

#Run the install script for NVIDIA driver

sudo sh NVIDIA-Linux-x86_64-384.111.run --no-opengl-files

#If --no-opengl-files is not specified above, you will run into a login loop.
#If you get the message that the script failed, choose "continue"
#If you are given the option to register the kernal module sources with DKMS, choose "Yes"
#If you are given the option to install the 32-bit compatibility libraries, choose "No"

#Run the install script for CUDA

sudo sh cuda_8.0.61_375.26_linux.run --no-opengl-libs

#If you are given the option to install the NVIDIA driver, choose "no"
#Give the preferred location for installation or accept the defaults

#Reboot the system by typing

sudo reboot

#Add the cuda-8.0/bin location to the PATH
#Add the cuda-8.0/lib64 location to the LD_LIBRARY_PATH

Installing PLUMED

#Run the following commands for needed software

sudo apt-get install git cmake gawk
sudo apt-get install libmatheval-dev
sudo apt-get install libopenmpi-dev openmpi-bin
sudo apt-get install gnuplot
sudo apt-get install python3 python3-numpy python3-scipy

#Get the location of matheval.h which will later be specified with CPPFLAGS

sudo find / -name matheval.h

#Download PLUMED from http://www.plumed.org/get-it
#Untar the file and do the installation

tar -xvzf plumed-2.4.0.tgz
cd plumed-2.4.0
sudo ./configure CPPFLAGS=-I/usr/include CC=mpicc CXX=mpicxx
make -j 4
sudo make install

GROMACS installation

#Download gromacs 5.1.4 from http://manual.gromacs.org/documentation/5.1.4/download.html
#Untar, patch with PLUMED and install

tar -xvzf gromacs-5.1.4.tar.gz
cd gromacs-5.1.4
plumed patch -p --runtime -e gromacs-5.1.4
mkdir build
cd build

#Find the location of the nvml.h file and specify the directory path under -DNVML_INCLUDE_DIR later with camke

sudo find / -name nvml.h

#Find the location of the libnvidia-ml.so file and specify the file path under -DNVML_LIBRARY later with camke

sudo find / -name libnvidia-ml.so

#Type the below command in the same line.
cmake .. -DGMX_BUILD_OWN_FFTW=on -DGMX_MPI=on -DCMAKE_C_COMPILER=mpicc -DCMAKE_CXX_COMPILER=mpicxx -DGMX_GPU=on -DNVML_INCLUDE_DIR=/usr/local/cuda-8.0/include -DNVML_LIBRARY=/usr/lib/x86_64-linux-gnu/libnvidia-ml.so
make -j 4
sudo make install

This Post Has 8 Comments

    1. Sajeewa 10 Nov 2020 Reply

      You are welcome!

  2. Elizabeth 23 Oct 2020 Reply

    Thanks designed for sharing such a good idea, paragraph is fastidious, thats why i have
    read it entirely

    1. Sajeewa 10 Nov 2020 Reply

      You are welcome!

  3. Gustavo 1 Dec 2020 Reply

    Is this valid for Debian distribution?

    1. Sajeewa 20 Dec 2020 Reply

      It worked for me on Ubuntu. It would probably work on Debian as well. I haven’t tried though.

  4. Nikhil Barua 16 Mar 2021 Reply

    Did you install Ubuntu VDI in your local windows operating system or did you install Ubuntu operating system separately on your hardware?

    1. Sajeewa 17 Jun 2021 Reply

      This is for a machine running on Ubuntu only.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.