In order to achieve the objectives, the methodology described above addresses three main lines of action:
- Reducing the energy consumed by the simulation by at least an order of magnitude through the optimal use of low-level HW resources (accelerators), minimization of the data movement and optimal load distribution between compute devices.
The second quantifiable objective of NextSim is to divide Time per iteration of current simulations by a factor of F=50x, considering a configuration with the same number of cores than before but in an advanced processing architecture (many-core, GP-GP) framework.
Four complementary points of views will be checked in NextSim:
- Efficiency. This is the main objective. For a given problem (mesh) the computational time will be reduced.
- Insight. Higher and more complex meshes will provide better insight into the physics of the problems and the solution can be obtained in the same computational time as before.
- Accuracy. Extreme scaling implies variable, unstable, non-guaranteed, dynamic resources and this will introduce system noise. The effects will be quantified, introducing a “simulation error bar” concept to assess the reliability of the whole process.
- Reliability and robustness. Necessary for the final deployment of this technology at industrial level. It is necessary to trust on the obtained solution and it is necessary a small sensitivity under external perturbation or modifications of the flow management and parameters.
This project has received funding from the European High-Performance Computing Joint Undertaking Joint Undertaking (JU) under grant agreement No 956104. The JU receives support from the European Union’s Horizon 2020 research and innovation programme and Spain, France, Germany.