Dr. Moritz Lehmann, Computational Fluid Dynamics software developer FluidX3D, shared a visualization of the solution to one of the most difficult problems of this kind – the calculation of air flows around a supersonic passenger plane Aérospatiale-BAC Concorde on landing.
Computations were performed on the GigaIO SuperNODE GPU server equipped with 32 AMD Instinct MI210 GPUs with 64 GB and a total volume of video memory of 2 TB.
The calculation task is the simulation of air flows for 1 second during the flight of the Aérospatiale-BAC Concorde at a speed of 300 km/h with an angle of attack of 10°. The Reynolds number (a characteristic number in hydrodynamics based on the ratio of the inertia of a gas, liquid or plasma flow to its viscosity) based on the wingspan is 146 million (that’s a lot).
The simulation resolution is 2976 × 8936 × 1489 = 40 billion blocks, with a block size of only 12.4 mm³. 67,268 simulation steps were computed in 29 hours, plus 4 hours spent rendering 5 × 600 frames (five different angles) at 4K resolution, for a total of 33 hours on the GigaIO SuperNODE server. One render frame is based on 475 GB of data, so 600 frames is 285 TB of data.
The entire simulation is a test of the recently added free-slip algorithms at FluidX3D at object boundaries, which allow for a more accurate model for the turbulent air/fluid boundary layer.
According to the author, on the same hardware, some commercial computational fluid dynamics programs, such as Ansys or Star-CCM+, take several years for similar simulations. FluidX3D does it in a few days.
And, well, in general, it is very beautiful and helps us remember why powerful computers are actually needed.
View source version on mezha.media: https://mezha.media/en/2023/08/03/one-of-the-most-difficult-problems-of-computational-fluid-dynamics-concorde-during-landing-video/