Installing GPU Driver

GPU driver and CUDA Toolkit

1. Install GPU driver using GUI

   GPU driver can be installed via other methods too but this is the easiest.

   • Go to “Software & Updates”, go to “Additional Drivers” tab.
   • Select the latest driver. In this case, nvidia-driver-470(proprietary, tested)
   • “Apply changes” and Reboot.

   

• Open the terminal application and type nvidia-smi to see GPU info:

Here, as you can see the recommended CUDA version is 11.4. Download it from https://developer.nvidia.com/cuda-toolkit-archive.

• For installation in Ubuntu: select Linux > x86_64 > Ubuntu

Then you’ll see that it’s only available for Ubuntu 18.04 and 20.04. If the you’ve installed some other version of ubuntu, reinstall one of this, and repeat the whole process.

Installation instructions are given there. Just run those commands.

wget https://developer.download.nvidia.com/compute/cuda/11.4.0/local_installers/cuda_11.4.0_470.42.01_linux.run
sudo sh cuda_11.4.0_470.42.01_linux.run

• During the installation of the CUDA toolkit, it will tell that you have already installed a driver, bla bla bla. Select “continue”.
• Untick the driver. and continue with the installation.
• Open the terminal application and type nvcc -V
• It will show some error and recommend a command sudo apt install cuda-toolkit or something like that. Run that command.

• retype nvcc -V. And you’ll see that CUDA toolkit is now installed.

  

And voila, it’s done. (If not, just reboot)

LAMMPS with GPU

Everything is almost the same as regular installation. While building the lammps just change this command.

cmake -D PKG_GPU=on -D CUDA_ARCH=sm_30 -D CMAKE_CXX_COMPILER=g++ -D CMAKE_C_COMPILER=gcc -D CMAKE_INSTALL_PREFIX=/home/'computer_name'/lammps/lammps-install ../cmake 

run the input file:

lmp -sf gpu -pk gpu 0 omp 40 -in in.lammps

where, 40 = number of processors

in.lammps = input file

gpu 0 will use all the available gpus

You’ll also need to make some change in the input script. For that, read the lammps manual.