# NVIDIA’s Tech: Brutal 2,500,000 Part Simulation!

## Метаданные

- **Канал:** Two Minute Papers
- **YouTube:** https://www.youtube.com/watch?v=wq8BgIfOxnk
- **Дата:** 20.04.2025
- **Длительность:** 8:42
- **Просмотры:** 80,781

## Описание

❤️ Check out Vast.ai and run DeepSeek or any AI project: https://vast.ai/papers 

📝 The papers are available here:
https://www.dgp.toronto.edu/projects/trading-spaces/
https://pcs-sim.github.io/pd/
https://visualcomputing.ist.ac.at/publications/2024/SDTF/
https://starryuniv.cn/files/sig24magnetic.pdf
https://github.com/Univstar/IoB-Ferrofluid-2D

📝 My paper on simulations that look almost like reality is available for free here:
https://rdcu.be/cWPfD 

Or this is the orig. Nature Physics link with clickable citations:
https://www.nature.com/articles/s41567-022-01788-5

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Benji Rabhan, B Shang, Christian Ahlin, Gordon Child, John Le, Juan Benet, Kyle Davis, Loyal Alchemist, Lukas Biewald, Michael Tedder, Owen Skarpness, Richard Sundvall, Steef, Taras Bobrovytsky, Thomas Krcmar, Tybie Fitzhugh, Ueli GallizziIf you wish to appear here or pick up other perks, click here: https://www.patreon.com/TwoMinutePapers

My research: https://cg.tuwien.ac.at/~zsolnai/
X/Twitter: https://twitter.com/twominutepapers
Thumbnail design: Felícia Zsolnai-Fehér - http://felicia.hu

#nvidia

## Содержание

### [0:00](https://www.youtube.com/watch?v=wq8BgIfOxnk) Segment 1 (00:00 - 05:00)

This is an amazing paper that is going to be absolutely brutal. Yikes. I kind of love it. So, what is going on here? You see, most simulations in computer games are about simulating solids. That is the boring stuff. But what about when we try to destroy things and they deform? And then we will do increasingly crazier things like this and then this and at the end this insane thing. and then I'll tell you why I am heartbroken. Dear fellow scholars, this is two minute papers with Dr. Koa Eher H. Yes, now I see the appeal of simulating defamations. That would be amazing. But doing this on a larger scale that is so much more difficult and takes much, much longer. Dropping a spiky maze on this city is a beauty. But some of these simulations can take 3 hours to compute or even longer. Wow, that is brutal. I mean, the simulation and the waiting time, too. But wait, maybe this new paper can help us. Although, I doubt it because it would take simulating 2 and a half million tetrahedra, these tiny little elements. That sounds very painful. Like this painful. However, it can compute a simulation. Once again, kind of brutal. And surprisingly, it gets even better. Don't forget this is a virtual world, our world. So what does that mean? It means we do whatever we want. We can control the stiffness of these objects with just one physical parameter. Just change it. And then look, we just made that jelly a bit more rubberized. Fantastic. Do it some more. And now the anvil barely bounces off of it. Loving it. So let's pop the question. Big breath. So, how fast is it? Now, hold on to your papers, fellow scholars, because when I saw this, my goodness, I couldn't believe my eyes. It is between 30 to 300 times faster than previous methods. Some of the smaller simulations take only a few seconds. So, perhaps just one more paper down the line and we might just get this in real time in our video games. Loving it. Now, when talking about class simulations, just look at that. This other work really knows how to do these really tough twisty cases. But it gets better. And not just because it can simulate the flag of California flawlessly. Well done there. But typically when we want to achieve something in a computer game or animated movie, we can't compute a full scale simulation like this because it can take from hours to days to compute. No, no. First, we compute a course simulation quickly. See if it has promise. But o, we have a huge problem here. We can't do this. Do you see why? making a finer version of a core simulation behaves entirely differently. So, we have to wait for days for the final simulation and that is not what I want. I mean, having to wait so long to get a chance to throw it a bit differently again, no thanks. But with this incredible method, yes, that's what I want. A perfectly designed experiment. We do the coarse experiment quickly and clear all three rings finally. And normally we saw that when we do the finer version of the simulation, something entirely different happens. And in this case, I can't believe it. Fast previewing of a difficult simulation is now possible and the outcome remains the same when running the full workload afterwards. I've never seen anything like this before. Bravo. So, if you want to make cats kiss or have these octoats fall into their own containers, you can simulate that very, very quickly and only do the fine simulation once afterwards. Yes, these are us fellow scholars when seeing this amazing paper at work. And while computer games still have lots of problems like this that we call Z fighting, where two seemingly simple objects just can't decide who should be in front. But look at this. Holy mother of papers. In the meantime, scientists are doing their best. And goodness, modeling crazy topology changes with these beautiful bubbles is also now possible. Just look at that. And imagine sitting down and having to write a handcrafted computer program to be able to do all that. This ingenuity is humanity at its best. Or with this

### [5:00](https://www.youtube.com/watch?v=wq8BgIfOxnk&t=300s) Segment 2 (05:00 - 08:00)

you can also stack a bunch of objects together in twisty ways. And when you look inside, let's see. I don't see any fighting at all. Fantastic. And I kept my favorite for last. Oh my. Are you seeing what I am seeing? Simulating a piece of fluid that is magnetic ferrofluids. It is very simple except the fact that it is close to impossible. Let me explain. All you need to do is put a magnet under a piece of ferrofluid and these magical spikes start appearing. That is simple. Now sit down and write a computer program that is capable of simulating that. Now that fellow scholars is nearly impossible. Yes, you have to understand and program all this crazy stuff to be able to pull this off. So what is going on here? Well, this work offers something they call an induce on boundary solver. What it can do is that it does not perform the computations on the entire 3D volume of the fluid, only on the two-dimensional surface compute on the shell. That is much quicker. And they pulled it off in a way that offers more favorable computational speeds than previous works and can be dropped into an existing fluid simulator. And this is how you can create these amazing fluid mazes. and other insane experiments. I love these works so much. That is a problem. I will tell you about it in a moment. Yes, you still have to wait for quite a while, but you know, for this kind of quality, I'll let it slip. And you know, everyone talks AI this, AI that. But I see AI as a tool to enhance the minds of these incredibly brilliant researchers. Just imagine what we will be capable of just two more papers down the line. What a time to be alive. And now look at this. My heart is broken as almost nobody is seeing or talking about these amazing papers. Can you believe that? Here on Two Minute Papers, you can learn about works often no one else is talking about. But here's the problem. It is almost impossible to keep the flame alive for simulation papers like this. You see, a few hundred episodes ago, we had ones that did really well, and YouTube kept recommending these episodes. And I get it. I mean, look at this insane quality work. Wow. But unfortunately, YouTube is not recommending them to you too much anymore. So, every time I am just here talking to myself whenever I do that, probably this time, too. Hello, Caro. Hello. How are you doing? Doing great, thank you. Except that it's been almost a thousand 2-minute papers videos, God is my witness. I tried everything since. But if it's a simulation paper, nothing works. I am heartbroken. I don't really have a solution. But if you keep watching these, posting them, and recommending them to your friends, maybe one day, maybe. So, thank you for being with me on this journey for almost 10 years now. This is my dream job and we couldn't exist without you fellow scholars. Let me know in the comments what you think. And if you wish to run Deep Seek yourself, Vest AI helps you rent Nvidia GPUs to do that for an incredible price. I love it because my costs come down to a fraction of a dollar for a session. It is incredible. Make sure to try it out today at vast. ai/papers AI/papers or click the link in the description.

---
*Источник: https://ekstraktznaniy.ru/video/12443*