Note: This article originally appeared on tomshardware.com, click here to read the original piece.
By Mark Tyson
GigaIO’s SuperNODE rendered a FluidX3D CFD simulation at an astounding 40-billion-cells resolution.
One of the largest Computational Fluid Dynamics (CFD) simulations ever was rendered over the weekend. The CFD simulation of just one second of Concorde landing at 300 km/h rendered at a resolution of 40 billion cells (2976×8936×1489) and took 33 hours to process. This single second of simulation is said to take up to nine years using conventional CFD programs.
Dr. Moritz Lehmann, AKA @ProjectPhysX, conducted the simulations with the FluidX3D program on GigaIO’s SuperNODE, which packs 32x AMD Instinct MI210 GPUs with a total of 2TB of VRAM.
We also have a few other entertaining videos of Lehmann’s other CFD projects below, like an armchair simulation on $45,000 worth of GPUs, and the aerodynamics of a cow. Yes, a cow.
In the video above, you can see the iconic Anglo/French supersonic passenger jet with its flaps down at a 10° angle of approach, causing a lot of turbulence during descent and deceleration. This CFD shows what is happening to the airflow from a multitude of angles. From some views, it looks like the air is boiling off the Concorde’s carefully crafted lines.
Dr. Lehmann shared more technical details about the Concorde CFD in a Reddit thread. Astonishingly, he said each frame of the simulation visualizes 475 GB of volumetric data, so the full minute takes up 285 TB total.
According to Dr. Lehmann, a “commercial CFD would need years for this, [but] FluidX3D does it over the weekend.” Dr. Lehmann is actually the sole developer of FluidX3D, as well as a Khronos OpenCL Advisor. Living on the cutting edge of CFDs and graphics, he had the foresight to develop FluidX3D (OpenCL) with 32-GPU scaling ‘out-of-the-box.’
If you want to delve even deeper, Dr. Lehmann has shared the SuperNODE AMD Instinct GPU benchmarks and FluidX3D source code on GitHub. The FluidX3D software is free for non-commercial use.
Armchairs and Cows Fly, and Other Lehmann ProjectPhysX CFD Projects
The above-detailed Concorde CFD is by no means the first such project shared by ProjectPhysX on social media channels. Before he got his hands on the GigaIO’s SuperNODE, Dr. Lehman practiced his art on more modest systems.
In the example above, you can see what is claimed to be “the most detailed CFD simulation of a quadcopter ever,” with a resolution of 3 billion cells. The embedded video was rendered using FluidX3D, again, but this time on a quartet of Nvidia A100 40 GB GPUs.
If quadcopters are too normal for you, what about a CFD simulation for an armchair at 50 km/h?
The above armchair video was again rendered on the 4x Nvidia A100 40GB GPU system. Armchairs are designed for comfort rather than their aerodynamic qualities, and you can see a lot of drag due to ‘chairodynamics’ in action.
View source version on tomshardware.com: https://www.tomshardware.com/news/32-amd-mi210-gpus-render-massive-simulation-33-hours-for-one-second-of-concorde-landing