Finally, A Blazing Fast Fluid Simulator! 🌊
Вставка
- Опубліковано 29 вер 2024
- ❤️ Check out Weights & Biases and sign up for a free demo here: www.wandb.com/...
The shown blog post is available here:
www.wandb.com/...
📝 The paper "Fast Fluid Simulations with Sparse Volumes on the GPU" and some code samples are available here:
- people.csail.m...
- www.researchga...
📸 Our Instagram page is available here: / twominutepapers
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Alex Haro, Alex Paden, Andrew Melnychuk, Angelos Evripiotis, Benji Rabhan, Bruno Mikuš, Bryan Learn, Christian Ahlin, Daniel Hasegan, Eric Haddad, Eric Martel, Javier Bustamante, Lorin Atzberger, Lukas Biewald, Marcin Dukaczewski, Michael Albrecht, Nader S., Owen Campbell-Moore, Rob Rowe, Robin Graham, Steef, Sunil Kim, Taras Bobrovytsky, Thomas Krcmar, Torsten Reil, Tybie Fitzhugh
More info if you would like to appear here: / twominutepapers
Meet and discuss your ideas with other Fellow Scholars on the Two Minute Papers Discord: discordapp.com...
Károly Zsolnai-Fehér's links:
Instagram: / twominutepapers
Twitter: / twominutepapers
Web: cg.tuwien.ac.a... - Наука та технологія
"If no one watches them, so be it, I still love doing this"
You are too kind Bernhard. Thank you so much! 🙏
Agreed
Ai breakthroughs grab my attention more often than fluid simulation but little by little, Two Minute Papers has worn me down and now I get excited for those too.
@@TwoMinutePapers So are there any softwares actually using this technique youre talking about then? Very misleading title if you dont actually share where you got all of this from
594k ** now 3 days later ;D
"Dear fellow scholars..."
Yep, I'm most definitely a scholar. I spend most of my days scholaring with other scholars at the local scholartorium. I was just saying to my friend earlier today (also a scholar), wouldn't it be great if everyone were scholars? Anyway, great vid. Scholar out!
👌
XD
I went to school! That makes me a scholar right? *Right?* :(
dropped out and do not intend to study again, do I qualify because I'm a curious idiot still ?
@@sankhyohalder97 Maybe you have to go to a "scholl".
That reminds me I've forgotten to do my German exercises today!
what a time to be alive
but on another note; those ferrofluid simulations are amazing
Completely agreed - absolutely incredible work! If you wish to see more, check this out. - ua-cam.com/video/goD36hVVl7M/v-deo.html
@@TwoMinutePapers thanks! I had a look and wow; the paper and work in that is crazy, great stuff.
Magnetofluid dynamics are a heck of a thing
Well...THE time to be alive would be when all the physics and AI simulations show are mainstream happening at real time on videogames and such
@@estebansingh9411 I've been working on one with powered ragdolls in wind, and water that's part of the game mechanics, not just an effect, for 10 years as an experient, we called Kontrol and updated weekly for a while. Stopped. a few years back because of too few appreciated the challenge of wind effects and physics made it too hard to travel, and we had spent tens of thousands already just so creature didn't walk like a drunk. It's in 2d because we can see inside workings, there are allot of firsts, 3d oceans I predict in 10 to 20 yrs studios will do it more like kontrol does. THings like surfing on wind usign just air drag and friction, were discovered and I never felt greater thrill in my career than to see these things emerge, I tuned and shaped surfboards all night to see how they performed In
stead of coding the behaviour, and the custom sparse air model I used is nothing like what can be done with modern pc or phone. I firmly believe all aaa games will use physics for all movement, include air and pressure with balancing robot controllers that can fall. Its a time when playing games will make u better at sports, not worse. I want to add fluids because they are so fascinating, especially vortices. In a year or so I hope to release an android update with vids on my channel. It's rough but playable now somewhat.
Woooooow! I'm facing that problem exactly right now: I'm working on a simulation of an aircraft splashing down into water, and given the speed of the aircraft and the splash in all directions, the domain is huuuuge. At any vaguely reasonable resolution that means 32 GB of RAM fill up long before the first frame is calculated, the software freezes and no patience in the world will let the simulation finish... so this is gold!!
Love watching fluid simulations. Thanks for doing the work to get these videos up. It IS appreciated.
I am a graduate student in geology. Fluid simulation applied to geosciences would be an amazing tool to understand the dynamics of rivers, coastal environments, landslides, volcanoes and so much more! I can't wait to see it, and to try using these tools myself. With these faster algorithms we will soon be able to work with large scale modeling, of entire depositional environments. Amazing.
Nobody showed up to _fluids simulations_ videos? But that's one of the most fascinating (and satisfying to watch) kinds of physics simulation!
Can’t wait till this comes to blender
Have you heard of the "taichi elements" addon that is currently under development? I would say the "taichi" thing sounds pretty similar to this. Here is a link to a video about Taichi:
ua-cam.com/video/wKw8LMF3Djo/v-deo.html
also the github page for taichi elements addon:
github.com/taichi-dev/taichi_elements
you might as well take your blender license and blend it up in a real blender
@@sharantheboss2957 haha what
That mantaflow bro...
These kind of videos and thought experiments are important. Sharing them helps future scholars or even something as simple as curious children push the technology further in their future. Math, technology, and a push for progress do not happen in lonely, isolated environments. Thank you for sharing. Knowledge should never be lonely.
Fascinating how our understanding of math and laws of physics can reproduce a realistic result!
Not computing the empty spaces reminds me of the various optimizations that were made for The Game of Life, which is a famous cellular simulation.
The naive implementation was to browse each cell and apply the logic.
The optimizations were about splitting the space in quad trees, and compressing it so that all the empty spaces ended up in a single quad. And all similar cell arrangements would be also grouped in a single quad. Apply the logic on the quads, and it would be applied on all the similar arrangements! Hence saving computing the same logic on the same cell arrangements. Then you could split the space even more in bigger blocks to find bigger patterns and simulate these once. And so on.
I'm just imagining this implemented in video games. Developers wont have to work as hard to make beautiful scenes
1:40 THe Water update: We decided to make sure that you have the most reallistic water ever so we did it! In order for it to happen in real time we decided to remove everything but the water and block meshes.
last time i was this early i started a fluid simulation with an old method, its still not done yet.
im ready for the day when video games start to have real time fluid simulations like this
fluids are so cool. i remember an old nvidia demo that runs 40.000 particles at 60 fps some years ago. i bet a more modern demo could have way more. was really nice to look at with multiple render styles, like clear/differently colored water, oil or even liquid metal. fun times
It took Blender many years to incorporate FLIP fluid simulations… But let’s hope the open nature of this paper means it can be implemented sooner haha. Would LOVE to try this out
When you say "finally" I imagined it would be a paper that just came out, not from 2018.
I know - haha. And nearly 8 minutes long. FALSE ADVERTISING 😅
Still great, though.
@@jacanchaplais8083 You should ask for a refund, or pay 4 times as much. Same same.
want this in blender asap! What a wonderful time to be alive!
Have you heard of the "taichi elements" addon that is currently under development? I would say the "taichi" thing sounds pretty similar to this. Here is a link to a video about Taichi:
ua-cam.com/video/wKw8LMF3Djo/v-deo.html
also the github page for taichi elements addon:
github.com/taichi-dev/taichi_elements
In 2014 I got a crappy job at a crappy community college and my boss says what do you know about computational fluid dynamics? Well, I'd love to say that I started on a journey that led me to writing papers and being an expert on cfd but the truth is... this stuff is so far beyond me. The moment I saw there was calculus involved I was like.. nope. Thanks though.
That doesn't mean I can't appreciate fluid simulations. I have kept my eye on every generation. Watching the improvements. And I just love it.
Thanks for making these videos :D I enjoyed seeing new progress in this interesting field.
Really enjoying the fluids content. I'm working on reduced order modelling of neutronics, and it's great to see some related work in fluids that's making good progress!
THANKS TO YOU for trusting in yourself and what you like
I've been looking forward to a real-time nonlinear CFD and planning to design a feedback controller for UAVs using the technique. But such a real-time numerical N-S equation solver is nearly impossible for now. However, watching this video, I am convinced that the moment is not that far away. Nice work!!!
Very interesting, what kind of UAVs do you mean?
Don't know a thing about this subject, but god damn that was soo satisfying to watch
I love fluid Sims. Can't wait until I can use all this in Blender! Flip fluids is already awesome but my papers flew out of my hands with this one. 🤠
I have allways loved the fluid dynamics stuff, I think it's possibly how I originally found this channel.
Love the videos and passion behind them!
1:38 i nearly fucking lost it i thought he was actually going to say minecraft
One step closer to fully physics based water in games
Well, ofcourse we like fluids. After all we get all of our oxygen from a fluid
I don’t know what half the sentences you’re saying mean, but damn do I love it.
Truly.... What a time to be alive.
It would be nice to see videos of the types of artifacts that show up, like where you start seeing granularity or anisotropic artifacts. In the sense that getting to see how mpeg4 video generates very different artifacts from h.264 gives you an idea of the improvement h.264 brings over mpeg4 for a similar processing cost.
I know in the early cuda based simulations you could really seethe individual particles if you looked.
Is this doing something like decimating particles in low-complexity areas and increasing granularity in high-complexity areas as it runs the simulation? It seems like that would really compliment the dynamic bounding of the simulation space.
_Now_ things are looking eerily real.
I love the fluids
love it - thanks so much for posting I watch! I love it too!
We love the fluids, yes.
I love every single video! Thank you so much!!
Love the jam on the cream cake 😂👍😘😍🥰
Love your videos
Love your videos Doc! Stay safe and keep going with your flow!
I love how he calls us scholars when I'm just a high schooler who has absolutely no clue how any of this works
I'm no scholar, I just like to come in here and ruminate about how we might live in a simulation because we are already developing all the techniques to make a realistic simulation of physics. In just a few decades you will be able to go into virtual reality and it will be indistinguishable from real life.
oh my goodness what the heck, thank you for making this video.
Everytime he does the honey drip simulation.... Im like thats a faaatttt ass dab simulation. lol
ahh... siggraph. The tech demos that will almost never see themselves in gaming for a decade.
Fluid simulation is one of my favorite. Your summaries of all the papers are so concise and easy to understand.
I am interested in all the simulations you have shown in your channel and I am hoping to do something in relation to those, do you have any recommendations on how and where I can start(I am 17 years old, I am an absolute beginner in these computation research).
This would be incredible in a video game. I know this is super hardcore Computationally expensive, but it could be unreal.
Once ray tracing was expensive, next gen GPU should be able to do it in "real time" with only a few % performance loss
A really cool video today, one suggestion that I think would be super helpful and informative would be a 'time counter' overlayed on each video so it's easier to get an understanding of how long these simulations take to render.
Also, is there a 'standard machine' that you are referring to for how long these renders take? For example when it says 200ms/frame surely that will change hugely depending on the CPU (or GPU), so an overclocked 64 core Threadripper compared to my old Quad-Core won't both take 200ms per frame.
I also wonder if any of these new GPU accelerated simulations are using the Tensor Cores on Nvidia's cards which sound like they would be perfect for this stuff or if it just uses the general shaders?
As the old adage goes: "If you build it, nerds will come"
Don't worry we love fluids as much as you do ;)
Autodesk and SideFX will just take a research like this to be put into their pricey commercial softwares, without having to invent new stuff themselves. I wish Blender could adapt it faster than those big companies
Utah teapot FTW!
Every single time I here you say a video is sponsored by Weights and Biases, I think it's some running joke, but then I'm like "Oh, it is just an appropriately named company!"
I just discovered your channel, and it's pretty amazing
So, if we assume that Moore's Law (in terms of 'computing performance', assuming transistor counts translate rather directly) continues roughly as it has, we can run the 4M particle Water Drop simulation in realtime in a year or 4? If we're at ~170ms/frame now, and we would be targeting a 30FPS animation which is ~33ms/frame, we only need a 4x performance increase. Moore's law states that transistors double every 2 years.
These works, in combination with Raytracing, are going to revolutionize digital graphics.
Nigh-indistinguishable virtual worlds are coming, everyone! :)
[insta edit, math is hard. It's a 5x increase. Leaving rest of comment as it though]
I'm gonna be honest; everytime I see that a new video is out, my eyes lighten up like a little kid, as if it was christmas or something :)
Me: looks down at computer Computer: don’t even think about it
I know you like lighting simulations, but fluid are my favorite!
Keep posting.. amazing content.
that's amazing infomation!! thanks
we love The Fluids
I am very late to this, but the reason I think simulations are so gpu and cpu tanking for it to be frames per second and not frames per minute is because assets are way too big for simulations to run smoothly in video games. Those small buildings is proof that water can move very fast through the city when everything is small. When things are too big, it looks like the water is moving in slow motion when actuality it is moving in the speed that we needed it to be in real-time. Maybe simulations can be more simple if the the assets used for each paper is small?
I share all of the fluid simulation videos with friends.
Didn't understand a thing (I thought GPU were the rule for rendering), but it was a nice experience anyway
Thanks :D
Can't wait for video game graphics in 2040
1:38 *Damn nice Minecraft update.*
Oh god.. where am I going to find 30 kitchens?
All I want, is to see these used in games. Im still waiting like well over 10 years.
It would be so dope if this were to come to Blender.
Fluids is next raytracing!
You can combine both methods tho
the Tsunami should be a breaking wave or a peak (high point) overwhelming the land NOT a low leading edge. the other thing to add to the simulation would be the failure of the buildings since no high-rise is built to take the force of water more than half the height of the building could produce the bottoms would get knocked out. bridges have huge foundations. where as buildings use many smaller posts for vertical strength lateral pressure will make them fail
What about Nvidia Flex? It can run in real time, on the GPU, and can simulate fluids, cloth, elastic objects, smoke and fire.
Reverse engineering the matrix!
I can't wait until we can do fluid simulation in real time so we can use it in games.
why does this water look better than real life wtf
I feel like we’re just slowly discovering how our reality is simulated.
I’m only here for the satisfying simulation
For this to start being used someone has to make a unity package for this tech
.....mmh ? What ? Oh, I'm sorry, hehehe... I was just looking at that jello CGI simulation... dear God...
0:50 I have been waiting for this for years. AI simulate matter better each year. Future AI will create Artificial Reality which no body talk about.
more fluids pls i like
1:24 is it just me or can I HEAR the cube land
Yes, I can hear it to, so as the 80% of the Earth human population. It's called sinestezia. When you see something so impactful, so noisy, that you can hear it.
I just felt like telling it.
@@marcelljozsa6618 so close
@Marcell Józsa i knew someone was gonna smart me like that but I already saw the video xd
i just realized, what if we unload stationary objects untilmoving ones get close
Imagine this in a game
Get this into Blender, asap!
What is the average buffering time inbetween you presenting these papers to us and it becoming a tool widely available to the public?
Great Video
me watching knowing absolutely nothing about that he's talking about: Hmm yes yes I agree 🤔
I can't wait until this is in The Elder Scrolls haha 🤤
Sooo where is the fluid solver? I want to use it for a fluid simulation in Blender.
looks so coool
I saw Flip Fluids being used in some of these graphics. Is this video an example of how Blender works Fluid Sims with Manta Flow?
At first I thought it said fast food simulator 😂😂
Your problem with working with particles is the space they are held in and the particles them selves do not have laws the space just holds the particles so you have to calculate each particles movement. How do the particles move and react to objects without counting and priority cues, how do they act on their own. That might need new hardware. It is a problem god solved with 4d space and atoms, it should be easy enough to do on a much smaller scale digitally. The laws of physics are pretty simpleminded, you just need the atoms or digital representations of atoms to do the work. A type of memory for physical space with more complex switches than on and off. Each position needs to store type of matter, speed of atom, whether atom is still or passing through, evolving direction and each grid layer, row or column of memory needs to measure weight of atoms on top of that layer and whether there is emptiness available for falling.
The water needs more weight to it, that city wouldn't be getting splashed wave heigh of 40 stories above the tallest building to put it in difficult perspective. Its splishing and splashing to high, and forcing upward at a high rate which simply isn't realistic.
isnt that what blender has had for a long time now? just turn on adaptive domain and optix acceleration, or Cuda. Isnt that the same?
Doctor, how can we be able to do it our computer? Will you please make a video on how to do these awesome fluid simulations in blender?
Did I hear you say Dr. (I won't embarrass myself trying to spell your name with my North American English keyboard)? Congratulations! Maybe I'm a little behind and this is old news but still, congrats.
Is the city from the tsunami simulation San Francisco?
Because i remember the piramid tower from GTA san andreas.
5 years and you'll see this stuff in new games. Someone call Unity, Epic and Nvidia pls!