Bull atos technologies and NVIDIA help three companies get more performance from their applications on GPUs

June 23, 2015

Forum Teratec – June 23, 2015 – In October 2014, NVIDIA and Bull, Atos Technologies together, launched a call for projects to help scientists accelerate their applications on NVIDIA GPUs. Scientists, researchers from public labs or innovative private companies were invited to submit their projects – and maybe win an assistance package to port their application to GPU accelerators.

Thanks to the support provided by Bull and NVIDIA, the winners will be able to port their code to GPU accelerators. Experts from the Center for Excellence in Parallel Programming will help them optimize and scale their existing GPU code, analyse and remove hot spots and tune their code on the latest generation of Tesla GPU accelerators.

Out of the many participating companies, three were chosen as exceptional examples of the benefits that can be derived from the massively parallel architecture of NVIDIA GPUs. The following three companies each received a free assistance package offered by Bull and NVIDIA: Dassault Aviation, the French aircraft manufacturer, Onera, the French aerospace research center, and Coria, a French research center with a focus on aerothermochemistry.

For Dassault Aviation, the accurate prediction of aircraft aeroelastic behavior (i.e. vibrations) is a requirement for aircraft design work. This is particularly the case in the transonic regime, where shock waves appear and disappear due to aircraft motions, and global aerodynamic properties can vary significantly. To tackle this problem, Dassault Aviation has developed and validated over the years a simulation-based strategy, which relies on AeTHER, its in-house Computational Fluid Dynamics software. Dassault Aviation hopes that GPUs will help shorten the run time of aeroelastic analyses, making them more readily available to aircraft designers.

The Onera project concerns the porting of an existing Fluid Dynamics Software for aeronautical flows. Their application allows engineers to simulate wakes for wings, fins, helicopter blades and helps to study the possible loss of lift* in flight. A 2D version of this GPU software has already been developed using the NVIDIA-powered ROMEO HPC center at the University of Champagne-Ardennes. ONERA now wants to focus on the scalability of the 3D version of their method for grids comprising billions of cells, for which Atos and NVIDIA expert advice would help.

*Lift is the force that allows an aircraft to rise and remain aloft.

Coria wants to optimize engine design, with a view especially to reduce the emission of pollutants. The Yales2 solver developed at CORIA is a research code designed to model turbulent combustion which occurs in many applications (aeronautical engines, automotive industry, industrial furnaces …). It aims at solving two-phase combustion from fuel atomization to pollutant prediction. The challenge will be to help CORIA to solve large linear systems (>1e9 Degrees Of Freedom) and ill-balanced problems on GPUs in order to accelerate them.

“The Bull Center for Excellence in Parallel Programming (CEPP) is very excited to contribute to these three projects and help them to utilize the performance of GPUs. Highly parallel system architectures offer a tremendous potential for research and innovation, provided applications can fully leverage their computing power” said Xavier Vigouroux, CEPP Manager, Big Data & Security, Atos.

“We are proud to leverage the expertise from Bull’s Center for Excellence in Parallel programming (CEPP) with the one from NVIDIA in order to bring these 3 strategic CFD applications to a new level of performance thanks to the NVIDIA GPU” added Walter Mundt-Blum, Vice-president for professional solutions EMEAI at NVIDIA.