Dr. Hamelin showed the seminar participants what EDF has achieved today and plans for tomorrow with examples. There are a lot of R&D challenges ahead within this energy company. EDF is largely involved with European research on HPC and simulation. It is a very technical and scientific involvement and a lot of money goes to research.

So why is there a need for HPC at EDF? There are two main applications, explained Dr. Hamelin. EDF has to manage very long term investments: plants last about 40-100 years in operation time. So, the company needs 3D simulation for design space to know where the limits are in energy production and delivery.

EDF evolved from 1D to 3D simulation in more than 10 years. More accurate simulation of cold water mixing in the core vessel adds 10 years to the life time margins. A decrease of the temperature in the vessel is the solution thanks to 3D simulation.

The EDF HPC challenge required faith in HPC, the speaker pointed out. It was a matter of bringing Tflops to the simulation floor. This had to trigger ideas and empower the EDF people to come up with solutions. Therefore the computer power was hugely increased.

A number of communities in R&D have to build on communities skills. There are 600 skilled simulation users and 150 developers. The company is enabling extensive simulation assets with 20 years of life time investment in the range of 100 million euro.

EDF is networking with academic partners. The HPC approaches involve detailed geometrics and tightly coupled physics. There are lots of uncertainties which call for parametric studies, Dr. Hamelin explained.

The earliest applications consisted of non destructive examinations. Very big systems are required, for example spider. First it was done with mock-ups.

We are talking here 150 billion photons with 1000 cores for calculation taking 24 hours. The R&D challenges for ultrasonic NDE are enormous since it is a 3D full way simulation which will have its completion in 2009.

Furthermore, nuclear fuel management studies are performed, as well as probabilistic LOCA studies with 74 uncertain parameters on 8000 cores, taking 12 days of calculation, according to the speaker. Flexible reloading involves 200.000 calculations and 10.000 cores.

Another problem studied consists of nuclear maintenance: which bolts can be affected by thermal and radiation degradation? This simulation requires 1,1 billion meshes and 2000 cores.

In a future not to far away, more power will be needed. HPC has to tackle the exponential complexity in simulation. Dr. Hamelin talked about 10 energy stocks, with 18 months of planning ahead. Today, the limit is 32.000 cores, which takes 1 hour of simulation.

EDF strives at a connection between academia and R&D people for a million core programming. This September there will be a Workshop with Jack Dongarra about simulation in the future and multi-scale materials modelling, announced Dr. Hamelin.

EDF is an end user with a number of opportunities, concluded Dr. Hamelin. It makes business sense to use HPC. The EDF connection to academic research in Europe is already active to advance knowledge and tools on many issues of common interest. Managing tier-0 HPC facilities for European research in a way that is connected to industry appears as a win-win approach.

The submission of industrial grand challenges on a scientific and HPC merit order basis adds to the variety of experiences for the whole academic community.

The uncertainty platform is open source and can be found at http://www.salome-platform.org