Large-Scale Fluid Simulations On Your Graphics Card | Two Minute Papers #123

Dear Fellow Scholars, this is Two Minute Papers with Károly Zsolnai-Fehér. The better fluid simulation techniques out there typically run on our graphical cards, which, if we formulate the problem in a way that the many compute units within can do in parallel, if they row in unison, if you will, we'll be greeted by an incredible bump in the speed of the simulation. This leads to amazingly detailed simulations, many of which you'll see in this footage. It's going to be really good! However, sometimes we have a simulation domain that is so large, it simply cannot be loaded into the memory of our graphical card. What about those problems? Well, the solution could be subdividing the problem into independent subdomains and solving them separately on multiple devices. Slice the problem up into smaller, more manageable pieces. Divide and conquer. But wait, we would just be pretending that these subdomains are independent, because in reality, they are clearly not, because there is a large amount of fluid flowing between them, and it takes quite a bit of algebraic wizardry to make sure that the information exchange between these devices happens correctly, and on time. But if we do it correctly, we can see our reward on the screen. Let's marvel at it together! Oh, hohoho, yeah! I cannot get tired of this. Typically, the simulation in the individual subdomains are computed on one or more separate graphical cards, and the administration on the intersecting interface takes place on the processor. The challenge of such a solution is that one has to be able to show that the solution of this problem formulation is equivalent to solving the huge original problem. And, it also has to be significantly more efficient to be useful for projects in the industry. The paper is one of the finest pieces of craftsmanship I've seen lately, the link is available in the video description. Thanks for watching and for your generous support, and I'll see you next time!