# NVIDIA’s New AI: Beautiful Simulations, Cheaper! 💨

## Метаданные

- **Канал:** Two Minute Papers
- **YouTube:** https://www.youtube.com/watch?v=bVxS9RXt2q8
- **Дата:** 21.09.2022
- **Длительность:** 6:33
- **Просмотры:** 271,299

## Описание

❤️ Check out Lambda here and sign up for their GPU Cloud: https://lambdalabs.com/papers

📝 The paper "NeuralVDB: High-resolution Sparse Volume Representation using Hierarchical Neural Networks" is available here:
https://developer.nvidia.com/rendering-technologies/neuralvdb
https://blogs.nvidia.com/blog/2022/08/09/neuralvdb-ai/
https://arxiv.org/abs/2208.04448

📝 The paper with the water simulation is available here: https://users.cg.tuwien.ac.at/zsolnai/gfx/photorealistic-material-editing/

❤️ Watch these videos in early access on our Patreon page or join us here on YouTube: 
- https://www.patreon.com/TwoMinutePapers
- https://www.youtube.com/channel/UCbfYPyITQ-7l4upoX8nvctg/join

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Aleksandr Mashrabov, Alex Balfanz, Alex Haro, Andrew Melnychuk, Benji Rabhan, Bryan Learn, B Shang, Christian Ahlin, Eric Martel, Geronimo Moralez, Gordon Child, Ivo Galic, Jace O'Brien, Jack Lukic, John Le, Jonas, Jonathan, Kenneth Davis, Klaus Busse, Kyle Davis, Lorin Atzberger, Lukas Biewald, Luke Dominique Warner, Matthew Allen Fisher, Michael Albrecht, Michael Tedder, Nevin Spoljaric, Nikhil Velpanur, Owen Campbell-Moore, Owen Skarpness, Rajarshi Nigam, Ramsey Elbasheer, Steef, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Timothy Sum Hon Mun, Torsten Reil, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here: https://www.patreon.com/TwoMinutePapers

Thumbnail background design: Felícia Zsolnai-Fehér - http://felicia.hu

Károly Zsolnai-Fehér's links:
Instagram: https://www.instagram.com/twominutepapers/
Twitter: https://twitter.com/twominutepapers
Web: https://cg.tuwien.ac.at/~zsolnai/

## Содержание

### [0:00](https://www.youtube.com/watch?v=bVxS9RXt2q8) Segment 1 (00:00 - 05:00)

Dear Fellow Scholars, this is Two Minute  Papers with Dr. Károly Zsolnai-Fehér. Today we are going to talk about an amazing  new technique to make these beautiful, but   very expensive computer simulations cheaper. How  much cheaper? Well, I will tell you in a moment. You see, around here, we discuss research  papers like this and this. We talk a great   deal about how difficult and computationally  expensive it is to simulate all this physics,   but that’s not the only thing that needs to  be done here. All this data also has to be   stored too! That is a ton of geometry  information. How much exactly? Well,   15 gigabytes for a simulation is not uncommon, and  my fluid simulation that you see here took well   over a hundred gigabytes of space to store.   Some of the more extreme examples can fill   several hard drives for only a minute of physics  simulation data. That is an insane amount of data. So, we need something to help compressing down all  this data. And the weapon of choice for working   with this kind of volumetric data is often a  tool called OpenVDB. It has this hierarchical   structure within, and scientists at NVIDIA had a  crazy idea. They said, how about exchanging this   with all of these amazing learning-based methods  that you are hearing about everywhere? Well,   okay, but why just throw a neural network at  the problem? Is this really going to help?   Well, maybe. You see, neural networks have  exceptional compression capabilities. For   instance, we have seen with their earlier paper  that such an AI can help us transmit video data   of us by, get this: only taking the first image  from the video, and just a tiny bit of extra data,   and they throw away the entire video afterwards! So, can these techniques do some more magic in   the area of physics simulations? Those are  much more complex, but who knows, let’s see! And now, hold on to your papers, because it  looks like this. Now that is all well and good,   but here is the kicker, it takes about 95% less  storage than the previous method. Yes, that is 20x   compression while the two pieces of footage still  look the same. Wow. That is absolutely incredible. And the previous OpenVDB solution  is not just for some hobby projects,   it is used in many of the blockbuster  feature-length movies you all know and   love. These movies won many-many academy  awards. And, we are not done yet. Not   even close - it gets even better! If you  think 20x is the best it can do, now look,   we also have 50x here. And sometimes, it  gets up to even 60x smaller. My goodness! And this new NeuralVDB technique from NVIDIA  can be used for not only these amazing movies,   but for everything else OpenVDB was  useful for. Oh yes! That means that   we can go beyond these movies, and use it for  scientific and even industrial visualizations. They also have the machinery to read and  process all this data with a graphics card,   therefore it not only decreases the  memory footprint of these techniques,   but it even improves how quickly they run. My  mind is officially blown. So much progress in   so little time. And this is my favorite kind  of research, and that is when we look into the   intersection of computer graphics and AI. That’s  where we find some of the most beautiful works. And don’t forget, these applications are  plentiful. This could also be used with   FourCastNet, which is a physics model that  can predict outlier weather events. The   coolest part about this work is that  it runs not in a data center anymore,   but today, it runs on just one NVIDIA  graphics card. And with NeuralVDB,   it will require even less memory to do  so. And it will help with all of these   absolute miracles of science that you see  here too. So, all of these could run with   an up to 60 times smaller memory footprint?   That is incredible. What a time to be alive! Huge congratulations to Doyub Kim, the first  author of the paper and his team for this amazing   achievement. And once again, you see the power of  AI, the power of the papers, and tech transfer at

### [5:00](https://www.youtube.com/watch?v=bVxS9RXt2q8&t=300s) Segment 2 (05:00 - 06:00)

the same time. And, since NVIDIA has an excellent  record in putting these tools into the hands of   all of us, for instance, they did it with the  previous incarnation of this technique, NanoVDB,   I am convinced that we are all going be able to  enjoy this amazing paper soon. How cool is that? So, does this get your mind going? What would you  use this for? Let me know in the comments below! Thanks for watching and for your generous  support, and I'll see you next time!

---
*Источник: https://ekstraktznaniy.ru/video/13441*