Efficient simulation of 1 billion particles

twelfth September 2022

Environment friendly simulation of 1 billion particles

By redesigning how fluids are simulated, researchers have demonstrated an order of magnitude pace improve on the earlier state-of-the-art for slow-flowing viscous liquids.

Modelling the behaviour of liquids is vital for a variety of functions, from industrial processes and medical units to pc graphics and visible simulations.

Nonetheless, regardless of a few years of improvement and identified physics courting again over 100 years, the power to precisely simulate liquid circulation stays some of the computationally difficult facets of making digital replicas of the true world.

It’s because the circulation and behavior of liquids is set by each the stress distribution by the liquid and, within the case of thick viscous liquids, the pressure-dependent inner resistance to that circulation. Exactly calculating these complicated and time-varying distributions could be very computationally intensive, and so many optimisation schemes have been developed to hurry up this calculation course of on the expense of accuracy.

Now, researchers on the King Abdullah College of Science and Expertise (KAUST) in Thuwal, Saudi Arabia, have made a major breakthrough in computational pace for viscous liquids, by combining environment friendly arithmetic with low-level parallel computational capabilities in fashionable pc processors.

“The simulation of fluid dynamics has been an evergreen matter in pc graphics analysis for a few years, and the prevailing strategies nonetheless have a lot potential for efficiency enchancment,” mentioned Han Shao, a PhD candidate at KAUST’s Visible Computing Middle. “On this analysis, we suggest the Unsmoothed Aggregation Algebraic Multigrid technique as a classy multigrid framework, which totally utilises fashionable CPU options and introduces new numerical strategies.”

Shao’s crew began with the concept extra environment friendly mathematical strategies might be used for the elemental matrix-vector calculations wanted to compute the stress distribution by the liquid, in addition to simplifying the calculation of trivial values on the fluid boundary.

Primarily, they confirmed that when many values within the matrices are the identical – as is the case within the bulk of a viscous liquid – a single calculation can be utilized throughout many components permitting many calculations to be skipped.

The crew then harnessed the synergy of mixing their environment friendly matrix-vector calculation code with the one instruction a number of knowledge (SIMD) functionality of contemporary CPUs, which permits the identical operation to be utilized to many knowledge inputs on the identical time. This enabled them to assemble a modelling method that might simulate viscous fluids with a billion particles as much as 15 occasions sooner than the present cutting-edge (the Houdini physics engine). In different phrases, a scene that after needed to be left working in a single day can now be totally rendered in solely an hour or so.

“Our framework can be utilized instantly by industrial customers for sooner simulation, utilizing the code accessible on our venture web site,” added Shao.