Installing QE and VASP (GPU version)

Comprehensive Guide: Installing Quantum Espresso and VASP (GPU Version) on Linux

This tutorial provides a detailed, step-by-step guide for installing Quantum Espresso (QE) and VASP (GPU version) on a Linux system using the NVIDIA HPC SDK. The instructions are based on the following versions:

• NVIDIA HPC-SDK: 23.5
• Quantum Espresso: qe-7.4.1 (tested with pw.x)
• VASP: 6.3.0

Note: Only pw.x from Quantum Espresso was tested, but the method should work for other subpackages as well.

---

Table of Contents

1. Prerequisites
2. Step 1: Install NVIDIA HPC SDK
3. Step 2: Install Quantum Espresso (GPU)
4. Step 3: Install FFTW Library
5. Step 4: Install VASP (GPU)
6. Troubleshooting & Tips
7. Conclusion

---

Prerequisites

• Linux system (tested on CentOS 7)
• Compatible NVIDIA GPU and drivers
• Basic terminal skills
• No root or sudo required

---

Step 1: Install NVIDIA HPC SDK

The NVIDIA HPC SDK provides the compilers and libraries needed for GPU-accelerated scientific computing.

Download and Extract

wget https://developer.download.nvidia.com/hpc-sdk/23.5/nvhpc_2023_235_Linux_x86_64_cuda_multi.tar.gz
tar xpzf nvhpc_2023_235_Linux_x86_64_cuda_multi.tar.gz
./nvhpc_2023_235_Linux_x86_64_cuda_multi/install

• Choose an installation directory, e.g., /home/youruser/nvidia/hpc_sdk.
• After installation, the SDK will be in /home/youruser/nvidia/hpc_sdk/Linux_x86_64/23.5.

Set Environment Variables

Add these lines to your ~/.bashrc or source them in your shell before building:

export PATH=/home/youruser/nvidia/hpc_sdk/Linux_x86_64/23.5/compilers/bin:$PATH
export MANPATH=$MANPATH:/home/youruser/nvidia/hpc_sdk/Linux_x86_64/23.5/compilers/man
export PATH=/home/youruser/nvidia/hpc_sdk/Linux_x86_64/23.5/comm_libs/mpi/bin:$PATH
export MANPATH=$MANPATH:/home/youruser/nvidia/hpc_sdk/Linux_x86_64/23.5/comm_libs/mpi/man

Or, if you use environment modules:

module load /home/youruser/nvidia/hpc_sdk/modulefiles/nvhpc/23.5

---

Step 2: Install Quantum Espresso (GPU)

Clone the QE Repository

git clone -b 'qe-7.4.1' --single-branch --depth 1 https://gitlab.com/QEF/q-e.git QEdir
cd QEdir
git submodule init
git submodule update

Prepare the Environment

• Check for Intel compilers:
  Ensure Intel compilers (e.g., ifort, mpiifort) are not in your path:

  which ifort
  # Should return: no ifort in (<paths>)
  

• Check CUDA capability and driver version:

  nvaccelinfo | grep -e 'Target' -e 'Driver'
  # Note the CUDA Driver Version and Default Target (e.g., cc70)
  

Configure and Build QE

Create a build script or run these commands manually:

#!/bin/bash
root_path=/home/youruser
cuda_runtime=12.1 # CUDA driver version
cc=70 # Compute capability

module load /home/youruser/nvidia/hpc_sdk/modulefiles/nvhpc/23.5

./configure --with-cuda=$root_path/nvidia/hpc_sdk/Linux_x86_64/23.5/cuda/$cuda_runtime \
            --with-cuda-cc=$cc --with-cuda-runtime=$cuda_runtime --enable-openmp

make -j 8 pw  # Builds pw.x using 8 cores
# make -j 8 all  # To build all QE executables

Running QE (pw.x)

Example PBS job script:

#!/bin/sh
#PBS -N test
#PBS -q batch
#PBS -l nodes=1:ppn=1
#PBS -l walltime=72:00:00

cd $PBS_O_WORKDIR
module load /home/youruser/nvidia/hpc_sdk/modulefiles/nvhpc/23.5

binpath=/home/youruser/Code/gpu/QEdir/bin

export OMP_NUM_THREADS=1
mpirun -n 4 $binpath/pw.x -i pw.in > pw.out

---

If you only need QE, you can stop here. For VASP, continue below.

---

Step 3: Install FFTW Library

FFTW is required for VASP.

cd /home/youruser/Code
wget https://fftw.org/fftw-3.3.10.tar.gz
tar xvf fftw-3.3.10.tar.gz
cd fftw-3.3.10

Build and install FFTW:

mkdir build
module load /home/youruser/nvidia/hpc_sdk/modulefiles/nvhpc/23.5
./configure --prefix=/home/youruser/Code/fftw-3.3.10/build
make
make check
make install
make installcheck

---

Step 4: Install VASP (GPU)

Prepare VASP Source

cd /home/youruser/Code
tar xvf vasp.6.3.0.tgz
cd vasp.6.3.0
cp arch/makefile.include.nvhpc_acc makefile.include

Edit makefile.include

• Set the FFTW path:

  FFTW ?= /home/youruser/Code/fftw-3.3.10/build # TOBECHANGED
  

• Set the CUDA version:
  Adjust the CUDA version to match your system (e.g., cuda12.1):

  FC  = mpif90 -acc -gpu=cc60,cc70,cc80,cuda12.1
  FCL = mpif90 -acc -gpu=cc60,cc70,cc80,cuda12.1 -c++libs
  

Build VASP

make all

If successful, you will find vasp_std, vasp_gam, and vasp_ncl in the bin directory.

---

Troubleshooting & Tips

• Compiler Conflicts:
  Ensure only NVIDIA compilers are in your PATH during build.
• Module System:
  If you use module load, ensure the correct modulefiles are loaded for each step.
• CUDA Version:
  Match the CUDA version in your system and in the HPC SDK.
• Permissions:
  Run commands as your user unless root is required.
• Documentation:
  Refer to the official NVIDIA HPC SDK Docs, Quantum Espresso Docs, and VASP Wiki.

---

Conclusion

You now have a working installation of Quantum Espresso and VASP (GPU version) using the NVIDIA HPC SDK. This setup enables high-performance, GPU-accelerated simulations for computational materials science.

If you encounter issues, consult the official documentation or search for solutions in user forums and communities.