GTC2013

GPU-accelerated Quantum ESPRESSO: Path to Petascale Simulations

| 24 October, 2012

Quantum ESPRESSO (QE) is an integrated suite of Open-Source computer codes for electronic-structure calculations and materials modeling at the nanoscale. It is based on density-functional theory, plane waves, and pseudopotentials. Quantum ESPRESSO stands for opEn Source Package for Research in Electronic Structure, Simulation, and Optimization. QE is evolving towards a distribution of independent and inter-operable codes in the spirit of an open-source project. Researchers active in the field of electronic-structure calculations are encouraged to participate in the project by contributing their own codes or by implementing their own ideas into existing codes.

A new build of the GPU-accelerated Quantum ESPRESSO version 5.0.1 (build2) is available for download. This new build includes few improvements:

  • - update of all CUDA kernels in order to get 10~15% more performance running new TESLA K10/K20 card
  • - phiGEMM 1.9.9
  • - MAGMA now compile and works on CRAY XK6 systems (using PGI + ACML 4.4.0)
  • - the recently introduced ELPA library (by adding manually __ELPA_PHIGEMM it is possible to couple ELPA and phiGEMM) very useful for large parallel gamma point calculations
This work has been supported by the PRACE 1st Implementation Phase project, WP 7.5e “Programming Techniques for High-Performance Applications – Accelerator”.  The Irish Centre for High-End Computing (ICHEC) has been directly responsible to extend the  Quantum ESPRESSO suite in order to support GPU calculation using the NVIDIA CUDA environment. Authors presented a paper focused on development work:
 
We  describe the different activities performed to enable the Quantum ESPRESSO user community to challenge frontiers of science running extreme computing simulation on European Tier-0 system of current and next generation. There main sections are described: 1) the improvement of parallelization efficiency on two DTF-based applications: Nuclear Magnetic Resonance (NMR) and EXact-eXchange (EXX) calculation; 2) introduction of innovative van der Waals interaction at the ab-initio level; 3) porting of PWscf code to hybrid system equipped with NVIDIA GPU technology.
 
Conclusion
The high complexity of supercomputing infrastructure expose scientific community to largely improve legacy code to efficiently perform large-scale simulation. This concept becomes even more strength if considering the introduction of accelerators on top of general purpose CPU systems. On the other hand the constant upgrade of complex scientific code requires to both developers and scientists a huge effort, large amount of computational resources and remarkable expertise in high performance computing.
 
The results presented in this paper proof that the model of strong collaboration between scientific communities and PRACE partners can overcome these problems by delivering a considerable contribution to tackle such challenges. Concentrating on selected communities a given mount of PRACE resources, we demonstrated that it is possible to enable complex scientific code to petascale Tier-0 system.
 
Other than highly scalable code, as a follow-up of this project, Prof. Stefano De Gironcoli submitted a proposal for PRACE production calls on Tier-0 system that has been accepted with high referee score at the first evaluation step. Final decision is till an on-going process at the time of writing. The GPU enabled version of PWscf has been the subject of many presentation, on-line articles and published on an IEEE conference proceeding.
 
Although the GPU version is still downloadable as a separated SVN branch all the work presented was recently released with the new Quantum ESPRESSO 5.0 version.
 
Filippo Spiga, Nicola Varini, Ivan Girotto (ICHEC, Ireland), Davide Ceresoli (CNR-STM, Italy), Carlo Cavazzoni (CINECA, Italy), Layla Martin-Samos (CNR-IOM, Italy – Univ. of Nova Gorica, Slovenia), Tommaso Gorni (Univ. Modena e Reggio Emilia, Italy) Enabling of Quantum ESPRESSO to Petascale Scientific Challenges PRACE White Paper, PRACE 1-IP, 2012 [Free PDF]
 

Tags: , , , , , ,

Category: Physical Science, Software