NVIDIA and IST Austria combined two water simulation techniques to capture both wake patterns and shore interactions in real time. The hybrid approach prevents water from disappearing, improves splashes and thin structures, and handles surface tension accurately in dynamic scenes like boats and moving containers.
This amazing paper is a collaboration between Nvidia and IST Austria on how to simulate water. But not just any piece of water. Like many, many previous techniques. No, no. These are large scale simulations with boats, dispersive reflections, waves sloshing onto the beach and oh my, a tiny bit of flooding as well. And you are going to hear about other papers from other groups like this and this. So back to floods and wakes. Is simulating that even possible? One previous technique is great at simulating wake patterns. These look great. However, look, on the shore the water behaves incorrectly. It essentially bounces back as if it hit a wall. This part can be simulated better with a shallow water simulation. But oh no, in that case the both wakes are completely off. Hmm. No one technique can perform both correctly. So what do you do then? Well, if no one technique works perfectly, you run both two theories. The really cool thing is that in this paper both are computed separately and then are combined into a beautiful unified simulation that can do everything correctly. That is absolutely incredible. And I would imagine probably takes forever more on that in a moment. But it gets better in three different ways. One is that even shallow water simulations are not perfect. Look, water can just disappear from the simulation over time due to numerical inaccuracies. So does the new technique solve that? Oh my, look at that. No more disappearing water. This is absolutely incredible because this new technique is not just fusing two methods together, but it is also making them better. Loving it now. It is better because two, the amount of detail in the simulations has also increased. Wow. There is a huge difference between the two. So more correctness and more detail. I bet this takes forever to compute. Probably takes a few minutes for each frame in this video. Right? Well, hold on to your papers, fellow scholars, because three, it all runs in real time. Goodness. And the vorticity generated around this obstacle is just mesmerizing. I mean, just look at that. And if you wonder how accurate it is. Well, get this. The paper states that it is a numerically perfect recreation of theoretical wave patterns. That is a strong claim that you don't see very often in papers like this. And yes, this is eye wateringly good quality. But there are also really exciting other works in this area. Well, not this one. You see, simulating fluid splashes where particles split and merge are really tough when we slow it down. Oh yes, you can see better. There is some goo here that refuses to split or oh my, even worse, it sticks to surfaces it should not stick to. And this new technique can. Let's see, now we're talking Fellow scholars, we get a much bigger splash at the back of the simulation domain. They are willing to separate more and are much less sticky. And we don't even need a huge splash for previous techniques to break down. If we decrease the resolution of the simulation, some droplets are willing to take the leap, while others not so much. And the new technique has perfectly willing particles. So good. But this previous ghost fluid method is perfectly fine when it comes to a damn break. Until you slow it down and look carefully. Do you see it? Oh yes. Some of the slits are completely clogged. This is not working well in the presence of thin structures. Yes, once again, thin structures. The two minute papers special. So how does the new technique perform? Look, the slits are now working correctly. Thank the papers. It can also do one of my favorite kinds of simulations. Yummy Surface tension for a cube in zero gravity, which, as it should, slowly becomes a sphere. They also did it with an armadillo, which is truly a sight to behold. Although the very end of the simulation seems to have been cut off. That's okay. I still love it. And having a good surface tension simulator doesn't only help with crazy experiments like this here. This is just a normal splash with no good surface tension added. Hmm. We see that something is wrong here, but it is hard to say what exactly. And with proper surface tension. Oh my. One more cohesive crown surface and a few droplets that say goodbye everyone. So good. Another fantastic paper. And finally, this other work shows how to simulate highly dynamic scenes where a liquid bunny is sloshing in a moving sphere. And in the meantime, it makes sure to have just the right amount of splashiness that previous methods failed to capture correctly. So I hope you enjoyed this. If you did, make sure to subscribe and hit the Bell icon and let me know in the comments below what kind of simulation you would make with these methods. I'd love to hear it. If you're looking for an Nvidia GPU cloud service for large scale AI projects, check out One Click Clusters by Lamda, featuring up to 512 Nvidia H100 GPUs with InfiniBand Networking. And hold on to your papers because with one click clusters you can access all these GPUs on demand with reservations starting at just two weeks. Try it out@lambdalabs.com paper or click the link in the description.
