"only for cut scenes, as realtime water effects at this quality and level of physics would kill any personal computer or console known to man."
So WTF we got AGEIA cards for ?? This is THE perfect opportunity to use them, granted somewhat simplified, but optimized properly surely as hell can get very close.
Actually, we can't do that in real-time yet. However, we _can_ get close, by turning down the resolution of the simulation. Each of those simulations are obviously using very fine liquid dynamics voxels, but by using larger voxels (volume pixels) you could get a similar "look and feel" in realtime. The water would look a little funny, but it would behave like water does.
So it IS possible now! As the state of the art grows better and better, our games would approach the full-reality appearance of the simulations you see here.
Same thing as full 3D games - the industry doesn't realize how badly it needs it... but it will.
"only for cut scenes, as realtime water effects at this quality and level of physics would kill any personal computer or console known to man."
Not necessarily. From what I gather, the videos where actual real-time, hence the phrase "physics simulations". And, if one could procure the money, the nVidia Quadro would probably handle it, coupled with about 4 gigs minimum of RAM. 'Course, this is only speculation.
Stunningly well done! This level of particle effects could also simulate many other real-world phenomena so well as to be utterly indistinguishable from the original, like blowing leaves, or pyroclastic flows, or bubbles rising, or cloud formation, or... heck, whatever hits your fancy.
Since I cannot create anything like this myself, I feel a little awkward bringing this up, but I do have a comment on the fire effect. The physics seem just a bit stiff compared to the real thing. For example, near the end of the first fire demo, the ball ignites with a dense smokey flame, but the smoke rises at much the same speed as the much hotter central column of flame, wholly unlike the real world. The outer sheath of cooler smoke would travel upward at a much lower speed, billowing outward, then curling (seemingly) downward, then being drawn inward by the far lower pressure of the faster moving central column.
But - I do not mean to belittle your results at all. Please let me say again what a fantastic job this is. Superior work. Pat yourselves on the back!
From what I gather, the videos where actual real-time, hence the phrase "physics simulations".
It's not rendered in real-time. The term physics simulation is to describe the process used to animate these videos. They used 'physics simulations', a set of constraints and coding, in order to produce the result that will be rendered at full quality.
From my experience with simulations, i would say they are almost never real time.
In the case of finite element simulation (simulating the goings on in a manufacturing plant) they are sped up drastically in order to predict work flow etc. (Ive done some ProModel work)
In the case of fluid dynamics they take forever due to the extremely complex interactions within the fluid. I did some modeling of aluminum casting of a small part and with a mesh which was very blocky looking, it still took my 2.4GHz machine a day or so to come up with 10 seconds or so of simulation. (I used solidcast / solidflow for this)
While Im sure these people use some very sophisticated software (solidflow is great but seemed to me a little dated) and some serious hardware (if it was blender, I would guess a cluster of some sort) this still took a long ass time to render.
As I see it, this is done as two separate processes: 1. the particle (physics) simulation 2. rendering the model (ray tracing)
Each of these takes *a lot* of computing. CPUs and GPUs of today can't even ray-trace in real time at some decent resolution.
I guess it took days (or more likely weeks) on some very powerful computers to create these (fairly low-res) videos.
So, what is the estimate (a *very* rough figure would do) of the processing power needed to simulate and render these in real time at some moderate (say 1024x768) full-screen resolution?
Then we can use this figure to speculate, how long will it take to get such power into a high-end PC (plus probably at least couple more years to get consoles of the generation following that).
That's when we can hope of seeing this in games (or other simulations of the environment)...
Having consoles do this in a gaming environment will take quite a bit longer. If you notice that the boundary are a of the simulation was quite small. The boundary areas in games are considerably bigger. Thus a lot more computing time to generate even approximations.
To the guy that said the fire wasn't entirely realistic, that might be true, but I think its fair to say game physics are never really true to real life and for good reason. We don't want to limit ourselves, games are meant to be creative, and this simulation looks as incredible (if not moreso) than any real flame.
I think the law of creating the best games physics comes down to one factor: how rad does it look? :P Real life just goes out the window when it comes to giant lasers, chainguns and rocket launchers.
Very impressive stuff. I think this could be useful in all manner of 3D work, not just videogames. I think games and other 3D materials all hope to be as close to reality as possible, but are usually limited by the power of their native systems or the expertise of their designers. These physics examples just bring fake reality one step closer.
For effects of this quality, we are looking at the simulation of several MILLION particles. If they were each 1 byte of information. The best processors would only provide 4gigabits of processing (in another words 250000particle simulation in realtime) and thats just simultation..
QUOTE: "Each of those simulations are obviously using very fine liquid dynamics voxels, but by using larger voxels (volume pixels) you could get a similar "look and feel" in realtime. The water would look a little funny, but it would behave like water does."
Actully, voxels have never been used for physics-based fluid simulation, since they don't behave correctly. It's all done on a particle-by-particle basis with inter-particulate forces governed by the effects artist (or as I'm guessing in this case, the student, since there's quite a bit more to be done before they're actually creating physically accurate water).
Voxels are often used to create physically accurate (or close to it) smoke and fire, however. I'm guessing the work seen here was a software R&D project, since much more accurate simulation tools already exist. Nevertheless, well done. I know I can't write my own simulation tools (which is why I use other people's).
Oh, and for all you gamers who expect this level of detail in the next big engine release, keep dreaming... It's nowhere to be found.
Voxels are used in nearly all professional fluid simulators in the FVX industry, including those involving liquids (gases are fluids too). The only professional level fluid simulator I know of that is particle based is RealFlow. Even blender's fluids dont use particles--it uses LBM last I checked. (Although, Nils Thurey's vid here looks like particles).
Most systems represent liquids with a "Level Set" volumetric isosurface--that is, instead of a density value on the grid, it is a signed distance value from the surface.
However, you are very right on one thing--this sort of thing simply does not happen in realtime. It can, and research has been done, but only at very very low resolutions.
One thing I GUARANTEE all of you--you will not see the fluid control concept in realtime anytime soon. I don't know how this one works, but the fluid control research I've read takes LOADS of memory and CPU time.
The fact of the matter is--these simulations sit in supercomputers for hours, even days. Most of the research in this field was strictly meant for the film industry, not the video game industry. Heck, Fedkiw got screen credit on Posiedon and Star Wars III.
There are better simulation tools out there ? Tell me where. Sure there is flowline from scanline and also ILM has a intern watersimulation tool. But for the most 3D Artists out there there is only Realflow wich sucks if it comes to big scenes. This is more a "water in the glas"-program. For fire there is even less for us 3D Artists. At the moment there is something in developement for max wich is called fume. But except this and maybe Afterburn for Max there is no real good Simulation Tool out on the market. And the videos with different fluids interacting was really great. The only tool I know wich is able to do this is flowline from scanline and that tool isn't avaible.
So instead of heading for realtime it would be great to release such a tool for 3D Artists first. This would make many commercials and little productions for TV so much easier and better.
Oh and I forgot to mention such simulations for games are still not possible. At the moment the cryengine seems to be the latest developement standart in this fiel and even this engine only works with prebaked simulations or if it comes to dynamic simulations than it only works with really few particles/objects (not over 100 as far as I have seen). So to think simulations like this (wich have to be calculated in realtime, you can't bake water simulations because they look different in every situation) won't be ingame in the next years.
"The fact of the matter is--these simulations sit in supercomputers for hours, even days."
Where do you guys get this stuff? You don't need a "supercomputer" to simulate these physics. They can be done on any halfway decent machine being sold today. The magic is in the programming. You gamers should do just that, play games and not speculate as to what it requires to simulate real world materials.
Yes, perhaps saying "supercomputer" was a bit of an exaggeration. However, these are ALL pretty short run, low resolution.
You need to approach simulation with the same attitude as rendering. Sure, you can run a simple ray traced scene at 480 24p in only ten seconds a frame on any half decent computer today. But if you want to render a 4K, 24p scene with full radiosity, 20+ pass antialiasing, caustics, and raytracing... forget it. You need a render farm.
Likewise, if you want to simulate a 100^3 gaseous fluid simulation, you can do that on any half decent machine today.
BUT, if you want to do a 500^3+ level set solution with gradient-based optimization for fluid control, high-iteration mass conservation, gap-filling projection (to have different fluids interact), chemical combustion, hi-resolution mesh generation, extremely high viscosity, viscoelasticity, vorticity confinement, DSP, "rigid fluid" (ie, two way solid fluid coupling as in the second video)... i could go on.
Expect to wait a LONG ASS TIME.
And where do I get this stuff? Uh, how about the 50+ publications I've read on the subject, and the personal experience I had building a realtime 2D fluid simulator java applet (based on Jos Stam's realtime code)?
And this is NOT "real world materials." Fluid simulation takes MASSIVE simplifications because it only needs to LOOK plausible--the fluid simulators in CFD/engineering fields are INSANELY complex, and those ones actually DO run on supercomputers.
Probably no realtime stuff the first few years in games, but it will get there eventially.
I remember seeing some fluids related stuff at nvidias site showing some apparently realtime fluids. However I don't know what system they used (or how many of their graphics cards they linked together to produce these results).
http://www.nvidia.com/object/geforce_tech_showcase.html it's the one in the bottom right corner: NVIDIA Quantum Effects™ Technology
Even a 4k rendering won't need to long on actual quadcores. Sure for a whole animation you probably would need some render slaves but as you know rendering is often also faking, most time you can get the look you want with rendertimes less than 30 minutes (for 4k maybe 2h but no one is rendering in 4k, not Pixar, not ILM...)
But from what I can see, we're already doing pretty much all the faking we can in fluid simulation--the NSE's these sims use are very much simplified already. Great leaps have been made using octrees and adaptive tetrahedral domains, but complex computations are nowhere near realtime. Only low res, basic simulations.
I guess what I'm trying to say is--this is pretty complicated stuff. People don't expect the graphics in video games to look like Shrek or The Incredibles, so likewise, this quality of fluid simulation just can't be expected in realtime.
That's not to say there isn't such thing realtime fluids--just don't expect cinema quality.
on a side note--no one renders at 4k? I thought most film was scanned at 4k. How about 2k? (I am admittedly much less knowledgeable regarding that stuff than fluid stuff...)
I can guarantee it is impossible to do these simulations at there resolution and quality at real time. Yes you can get moveing water from container to contain, and splashes and all that real time. But not at this quality. Also, Awsome job on the fire and ware. I just spent a week trying get fire down for a commercial and its a pain to get it just right. I didn't even bother getting into water lol. Anyways, Very impressive :)
For you people who say this kind of physics won't be in video-games in a long, long time, check out 'Hydrophobia'. It is maybe not as high quality simulation, but it's a good start. Homepage and Wikipedia
Anonymous said... I build these sims for a VFX Film company and they take days to render 1 at these voxel res. any home computer will blue screen in 5 mins.
--> wich one of the effects above ?. Even the water effects for poseidon where computet mostly on one single standart computer...
For you people who say this kind of physics won't be in video-games in a long, long time, check out 'Hydrophobia'. It is maybe not as high quality simulation, but it's a good start. Homepage and Wikipedia
--> The real question isn't if you can get such type of water ingame, the question is if you can simulate it real time. Like the cryengine works with huuuuge Simulations but this are prebaked simulations wich get resimulated ingame only for parts wich intersect with the simulation (like a tank wich stands in an explosion radius, there the simulation has to be resimulated for this part).
I couldn't find this informations on the 2 websites. They are talking about "real fluid dynamics" but not if they are presimulated or simulated in realtime.
"only for cut scenes, as realtime water effects at this quality and level of physics would kill any personal computer or console known to man."
One word for ya - BIOSHOCK.
Thanks for playing, please try again.
(btw, the processing power of most game platforms are rarely - if ever - pushed to their limitation by the developers. Just cuz you don't see the effect in game doesn't mean it isn't possible - just no one got creative enough to implement it... yet)
Great work! You almost had me fooled. However you lacked one important element in your simulations. You were missing how air and water interact with each other when water moves either through air or with it. You were missing bubbles.
Some people need to understand the PHYSICS going on here... Yeah, all kinds of amazing things can be done real time. But those are things that are designed to look as good as possible, while using an extremely simplified (or altogether different) internal mechanism. And as good as they get...and when you watch you can see that they are not as good as the real deal...
I've taken a couple graduate computational physics courses, so here are some real benchmarks for you.
200,000 particle simulation with only gravitational interactions (much simpler than anything that goes on in a fluid or flame simulation): 2-4 hours on a generic P4 processor. And that was just a baby thing that looked like crap. Full simulations are indeed usually ran on super computers. (But of course, that point is almost moot...200,000 particles wouldn't be nearly enough, but fluid dynamics isn't simulated using individual particles anyways)
30 seconds of fluid flow in a grid based solver using roughly a 1-10 thousand grids (can't remember off hand) ~1 hour, same processor as before. And again, its just a baby simulation. Nothing even close to the resolution you saw in any of these. And it was in a closed container. No waves. No surface tensions. No splashes and froth. Just raw flows/pressures/densities.
There are better codes out there than the ones I've been using. But not better enough to make order of magnitude differences. Serious physics simulations (for research purposes of course) are generally done on super computers. The videos here certainly don't need that kind of physics detail, and I'd bet were created on a general PC. But they most certainly need more than what I mentioned above, and so were not done real time. The above times were just for physics data creation, no graphics whatsoever involved.
Wow Love The Blog Comment from: Blackpool Hotels I must say remember seeing some fluids related stuff at nvidias site showing some apparently realtime fluids. However I don't know what system they used (or how many of their graphics cards they linked together to produce these results. Thanks Hotels Blackpool
To clear things up. All of these videos were simulated liquid physics (or liquid dynmaics). The 3rd one was from blender, but the others were from a high end program called Realflow 4.
Simulations are done in real time, BUT they are extreamly slow and CPU taxing. Its a VERY intense CPU process, not much so for RAM and video card. To get realtime results with fast performance and such we would need much more advanced and powerful CPUs than we have right now. So ya, in like 10 or so years this might be possible in games.
As for now... no an Agiea physx card woudnt do it, its for AGEIA physX supported games, few of..and is meant mainly for solid object and cloth physics. There wont be realistic looking real time water physics for a while.
btw u can check my site, ive done a few liquid simulations. www.freewebs.com/jsstudi0s
Wow! Really impressive!
Posted on June 14, 2007
Source is Ron Fedkiw and his students.
http://graphics.stanford.edu/%7Efedkiw/
Posted on June 14, 2007
Simply Awesome!!!!... I'd love to see those kind of effects in a Video Game... Wow just wow...
Kudos to the coders.
Posted on June 14, 2007
Well, you could try to design it by yourself too!
At least the third video was made with Blender, an open source 3d modeling, rendering, animation and videogame software!
www.blender.org
www.blendernation.com
Bye!
Posted on June 14, 2007
Come on people, let's get this shit in the video games quick! what's the damn hold up!?
Posted on June 14, 2007
"I'd love to see those kind of effects in a Video Game... Wow just wow..."
only for cut scenes, as realtime water effects at this quality and level of physics would kill any personal computer or console known to man.
Posted on June 14, 2007
Phenomenal work guys :)
Posted on June 14, 2007
"only for cut scenes, as realtime water effects at this quality and level of physics would kill any personal computer or console known to man."
So WTF we got AGEIA cards for ?? This is THE perfect opportunity to use them, granted somewhat simplified, but optimized properly surely as hell can get very close.
Posted on June 14, 2007
"surely as hell can get very close."
No they can't.. wait another 10 years though and we'll see.
Posted on June 14, 2007
Actually, we can't do that in real-time yet. However, we _can_ get close, by turning down the resolution of the simulation. Each of those simulations are obviously using very fine liquid dynamics voxels, but by using larger voxels (volume pixels) you could get a similar "look and feel" in realtime. The water would look a little funny, but it would behave like water does.
So it IS possible now! As the state of the art grows better and better, our games would approach the full-reality appearance of the simulations you see here.
Same thing as full 3D games - the industry doesn't realize how badly it needs it... but it will.
Posted on June 14, 2007
"only for cut scenes, as realtime water effects at this quality and level of physics would kill any personal computer or console known to man."
Not necessarily.
From what I gather, the videos where actual real-time, hence the phrase "physics simulations".
And, if one could procure the money, the nVidia Quadro would probably handle it, coupled with about 4 gigs minimum of RAM.
'Course, this is only speculation.
Posted on June 14, 2007
Stunningly well done! This level of particle effects could also simulate many other real-world phenomena so well as to be utterly indistinguishable from the original, like blowing leaves, or pyroclastic flows, or bubbles rising, or cloud formation, or... heck, whatever hits your fancy.
Since I cannot create anything like this myself, I feel a little awkward bringing this up, but I do have a comment on the fire effect. The physics seem just a bit stiff compared to the real thing. For example, near the end of the first fire demo, the ball ignites with a dense smokey flame, but the smoke rises at much the same speed as the much hotter central column of flame, wholly unlike the real world. The outer sheath of cooler smoke would travel upward at a much lower speed, billowing outward, then curling (seemingly) downward, then being drawn inward by the far lower pressure of the faster moving central column.
But - I do not mean to belittle your results at all. Please let me say again what a fantastic job this is. Superior work. Pat yourselves on the back!
Posted on June 14, 2007
From what I gather, the videos where actual real-time, hence the phrase "physics simulations".
It's not rendered in real-time. The term physics simulation is to describe the process used to animate these videos. They used 'physics simulations', a set of constraints and coding, in order to produce the result that will be rendered at full quality.
Posted on June 14, 2007
Was this done using ray tracing?
Posted on June 14, 2007
Would be nice to see some ocean waves.
Posted on June 14, 2007
do you guys just play video games to see the graphics of water and fire. thats really sad.
Posted on June 14, 2007
From my experience with simulations, i would say they are almost never real time.
In the case of finite element simulation (simulating the goings on in a manufacturing plant) they are sped up drastically in order to predict work flow etc. (Ive done some ProModel work)
In the case of fluid dynamics they take forever due to the extremely complex interactions within the fluid. I did some modeling of aluminum casting of a small part and with a mesh which was very blocky looking, it still took my 2.4GHz machine a day or so to come up with 10 seconds or so of simulation. (I used solidcast / solidflow for this)
While Im sure these people use some very sophisticated software (solidflow is great but seemed to me a little dated) and some serious hardware (if it was blender, I would guess a cluster of some sort) this still took a long ass time to render.
Impressive as hell.
Posted on June 14, 2007
Very good animations, however the final shot of "Fluid Physics: Water Control" is less impressive than the working shots.
Posted on June 15, 2007
Impressive indeed.
As I see it, this is done as two separate processes:
1. the particle (physics) simulation
2. rendering the model (ray tracing)
Each of these takes *a lot* of computing. CPUs and GPUs of today can't even ray-trace in real time at some decent resolution.
I guess it took days (or more likely weeks) on some very powerful computers to create these (fairly low-res) videos.
So, what is the estimate (a *very* rough figure would do) of the processing power needed to simulate and render these in real time at some moderate (say 1024x768) full-screen resolution?
Then we can use this figure to speculate, how long will it take to get such power into a high-end PC (plus probably at least couple more years to get consoles of the generation following that).
That's when we can hope of seeing this in games (or other simulations of the environment)...
Posted on June 15, 2007
Having consoles do this in a gaming environment will take quite a bit longer. If you notice that the boundary are a of the simulation was quite small. The boundary areas in games are considerably bigger. Thus a lot more computing time to generate even approximations.
Posted on June 15, 2007
To the guy that said the fire wasn't entirely realistic, that might be true, but I think its fair to say game physics are never really true to real life and for good reason. We don't want to limit ourselves, games are meant to be creative, and this simulation looks as incredible (if not moreso) than any real flame.
I think the law of creating the best games physics comes down to one factor: how rad does it look? :P Real life just goes out the window when it comes to giant lasers, chainguns and rocket launchers.
Posted on June 15, 2007
Your Aegia isn't going to help. While the simulation itself takes a while, the real killer is the raytracing for the refractions and caustics.
Posted on June 15, 2007
The splashes are looking like they have been inherited from mercury. Even if it looks much better than before, it has a long way to go
Posted on June 15, 2007
what's next? super robots that take over the world and kill's all humanity?
Posted on June 15, 2007
Check out some screenshots of Pirates of the XXI Century on www.openscenegraph.com
That water is even more impressive, with actual texturing being done in real-time, dynamically. And yah, it's 3d as well.
This stuff sucks in comparison.
Posted on June 15, 2007
Lots of ignorant people here. These are NOT REALTIME SIMULATIONS. They were RENDERED with a custom raytracer...NOT BLENDER.
Did you not check the source of the videos? Ron Fedkiw?
http://graphics.stanford.edu/%7Efedkiw/
Educate yourselves.
Posted on June 15, 2007
There's fire and water.. where's ice? :)
Posted on June 15, 2007
Very impressive stuff. I think this could be useful in all manner of 3D work, not just videogames. I think games and other 3D materials all hope to be as close to reality as possible, but are usually limited by the power of their native systems or the expertise of their designers. These physics examples just bring fake reality one step closer.
Posted on June 15, 2007
Such silly comments.
The generation of such effects comes in 3 parts.
Simulation of physical movement
Generation of mesh/geometry
Rendering of imagery.
For effects of this quality, we are looking at the simulation of several MILLION particles. If they were each 1 byte of information. The best processors would only provide 4gigabits of processing (in another words 250000particle simulation in realtime) and thats just simultation..
Realtime?? We are still a long way ahead.
Posted on June 15, 2007
QUOTE: "Each of those simulations are obviously using very fine liquid dynamics voxels, but by using larger voxels (volume pixels) you could get a similar "look and feel" in realtime. The water would look a little funny, but it would behave like water does."
Actully, voxels have never been used for physics-based fluid simulation, since they don't behave correctly. It's all done on a particle-by-particle basis with inter-particulate forces governed by the effects artist (or as I'm guessing in this case, the student, since there's quite a bit more to be done before they're actually creating physically accurate water).
Voxels are often used to create physically accurate (or close to it) smoke and fire, however. I'm guessing the work seen here was a software R&D project, since much more accurate simulation tools already exist. Nevertheless, well done. I know I can't write my own simulation tools (which is why I use other people's).
Oh, and for all you gamers who expect this level of detail in the next big engine release, keep dreaming... It's nowhere to be found.
Posted on June 15, 2007
NATHAN-- I assure you, you are quite wrong.
Voxels are used in nearly all professional fluid simulators in the FVX industry, including those involving liquids (gases are fluids too). The only professional level fluid simulator I know of that is particle based is RealFlow. Even blender's fluids dont use particles--it uses LBM last I checked. (Although, Nils Thurey's vid here looks like particles).
Most systems represent liquids with a "Level Set" volumetric isosurface--that is, instead of a density value on the grid, it is a signed distance value from the surface.
However, you are very right on one thing--this sort of thing simply does not happen in realtime. It can, and research has been done, but only at very very low resolutions.
One thing I GUARANTEE all of you--you will not see the fluid control concept in realtime anytime soon. I don't know how this one works, but the fluid control research I've read takes LOADS of memory and CPU time.
The fact of the matter is--these simulations sit in supercomputers for hours, even days. Most of the research in this field was strictly meant for the film industry, not the video game industry. Heck, Fedkiw got screen credit on Posiedon and Star Wars III.
Posted on June 16, 2007
reallly coooollllllllllll
Posted on June 16, 2007
I still have to pick up my jaw, wait...
Posted on June 16, 2007
@NATHAN
There are better simulation tools out there ? Tell me where. Sure there is flowline from scanline and also ILM has a intern watersimulation tool.
But for the most 3D Artists out there there is only Realflow wich sucks if it comes to big scenes. This is more a "water in the glas"-program.
For fire there is even less for us 3D Artists. At the moment there is something in developement for max wich is called fume. But except this and maybe Afterburn for Max there is no real good Simulation Tool out on the market.
And the videos with different fluids interacting was really great. The only tool I know wich is able to do this is flowline from scanline and that tool isn't avaible.
So instead of heading for realtime it would be great to release such a tool for 3D Artists first. This would make many commercials and little productions for TV so much easier and better.
Posted on June 16, 2007
Oh and I forgot to mention such simulations for games are still not possible. At the moment the cryengine seems to be the latest developement standart in this fiel and even this engine only works with prebaked simulations or if it comes to dynamic simulations than it only works with really few particles/objects (not over 100 as far as I have seen). So to think simulations like this (wich have to be calculated in realtime, you can't bake water simulations because they look different in every situation) won't be ingame in the next years.
Posted on June 16, 2007
"The fact of the matter is--these simulations sit in supercomputers for hours, even days."
Where do you guys get this stuff? You don't need a "supercomputer" to simulate these physics. They can be done on any halfway decent machine being sold today. The magic is in the programming.
You gamers should do just that, play games and not speculate as to what it requires to simulate real world materials.
Posted on June 16, 2007
Yes, perhaps saying "supercomputer" was a bit of an exaggeration. However, these are ALL pretty short run, low resolution.
You need to approach simulation with the same attitude as rendering. Sure, you can run a simple ray traced scene at 480 24p in only ten seconds a frame on any half decent computer today. But if you want to render a 4K, 24p scene with full radiosity, 20+ pass antialiasing, caustics, and raytracing... forget it. You need a render farm.
Likewise, if you want to simulate a 100^3 gaseous fluid simulation, you can do that on any half decent machine today.
BUT, if you want to do a 500^3+ level set solution with gradient-based optimization for fluid control, high-iteration mass conservation, gap-filling projection (to have different fluids interact), chemical combustion, hi-resolution mesh generation, extremely high viscosity, viscoelasticity, vorticity confinement, DSP, "rigid fluid" (ie, two way solid fluid coupling as in the second video)... i could go on.
Expect to wait a LONG ASS TIME.
And where do I get this stuff? Uh, how about the 50+ publications I've read on the subject, and the personal experience I had building a realtime 2D fluid simulator java applet (based on Jos Stam's realtime code)?
And this is NOT "real world materials." Fluid simulation takes MASSIVE simplifications because it only needs to LOOK plausible--the fluid simulators in CFD/engineering fields are INSANELY complex, and those ones actually DO run on supercomputers.
Posted on June 16, 2007
Very impressive stuff.
Probably no realtime stuff the first few years in games, but it will get there eventially.
I remember seeing some fluids related stuff at nvidias site showing some apparently realtime fluids. However I don't know what system they used (or how many of their graphics cards they linked together to produce these results).
http://www.nvidia.com/object/geforce_tech_showcase.html
it's the one in the bottom right corner: NVIDIA Quantum Effects™ Technology
Posted on June 17, 2007
Even a 4k rendering won't need to long on actual quadcores. Sure for a whole animation you probably would need some render slaves but as you know rendering is often also faking, most time you can get the look you want with rendertimes less than 30 minutes (for 4k maybe 2h but no one is rendering in 4k, not Pixar, not ILM...)
Posted on June 17, 2007
Thats true, a lot of it is faking.
But from what I can see, we're already doing pretty much all the faking we can in fluid simulation--the NSE's these sims use are very much simplified already. Great leaps have been made using octrees and adaptive tetrahedral domains, but complex computations are nowhere near realtime. Only low res, basic simulations.
I guess what I'm trying to say is--this is pretty complicated stuff. People don't expect the graphics in video games to look like Shrek or The Incredibles, so likewise, this quality of fluid simulation just can't be expected in realtime.
That's not to say there isn't such thing realtime fluids--just don't expect cinema quality.
on a side note--no one renders at 4k? I thought most film was scanned at 4k. How about 2k? (I am admittedly much less knowledgeable regarding that stuff than fluid stuff...)
Posted on June 17, 2007
Blog, you should not use the word ignorant too quickly. The third video is from http://graphics.ethz.ch/~thuereyn/ntoken3/Publications.html
His author has code the fluid simulator in Blender since 2.40 :
http://www.blender.org/development/release-logs/blender-240/fluids-simulation/
So, you can try fluid simulation and render in Blender.
Regards
Posted on June 18, 2007
Jep 2k is the actual standart for cinema. And more and more it's necessary to render in 2k for TV too.
Posted on June 18, 2007
Playstation 3 can do that.
Posted on June 19, 2007
Can do what ?
Posted on June 19, 2007
I build these sims for a VFX Film company and they take days to render 1 at these voxel res. any home computer will blue screen in 5 mins.
Posted on June 19, 2007
exactly, these can't be done in real-time ;). It has too many calculations to be done in real time (even on powerful CPUs and GPUs )
Posted on June 19, 2007
HydroEngine (Real time fluids)
http://www.bladeinteractive.com/hydroengine.html
Hydrophobia (HydroEngine Game)
http://www.bladeinteractive.com/hydro.html
Posted on June 20, 2007
Awesome!
Posted on June 20, 2007
I can guarantee it is impossible to do these simulations at there resolution and quality at real time. Yes you can get moveing water from container to contain, and splashes and all that real time. But not at this quality. Also, Awsome job on the fire and ware. I just spent a week trying get fire down for a commercial and its a pain to get it just right. I didn't even bother getting into water lol. Anyways, Very impressive :)
Posted on June 20, 2007
Great stuff!
Posted on June 21, 2007
For you people who say this kind of physics won't be in video-games in a long, long time, check out 'Hydrophobia'. It is maybe not as high quality simulation, but it's a good start. Homepage and Wikipedia
Posted on June 21, 2007
Anonymous said...
I build these sims for a VFX Film company and they take days to render 1 at these voxel res. any home computer will blue screen in 5 mins.
--> wich one of the effects above ?. Even the water effects for poseidon where computet mostly on one single standart computer...
For you people who say this kind of physics won't be in video-games in a long, long time, check out 'Hydrophobia'. It is maybe not as high quality simulation, but it's a good start. Homepage and Wikipedia
--> The real question isn't if you can get such type of water ingame, the question is if you can simulate it real time. Like the cryengine works with huuuuge Simulations but this are prebaked simulations wich get resimulated ingame only for parts wich intersect with the simulation (like a tank wich stands in an explosion radius, there the simulation has to be resimulated for this part).
I couldn't find this informations on the 2 websites. They are talking about "real fluid dynamics" but not if they are presimulated or simulated in realtime.
Posted on June 21, 2007
"only for cut scenes, as realtime water effects at this quality and level of physics would kill any personal computer or console known to man."
One word for ya - BIOSHOCK.
Thanks for playing, please try again.
(btw, the processing power of most game platforms are rarely - if ever - pushed to their limitation by the developers. Just cuz you don't see the effect in game doesn't mean it isn't possible - just no one got creative enough to implement it... yet)
Posted on June 22, 2007
Great work! You almost had me fooled. However you lacked one important element in your simulations. You were missing how air and water interact with each other when water moves either through air or with it. You were missing bubbles.
Posted on June 22, 2007
what??? that isnt possible.. is it??
Posted on June 25, 2007
what is the idea of this sh**?
Posted on June 25, 2007
THAT FLUID/FIRE SIMULATIONS
LOOK LIKE STANDARD MAYA FLUIDS TO ME!
Posted on June 26, 2007
also no!
this is not possible in realtime!!
so i cannot understand why this is posted on a gaming website!
Posted on June 26, 2007
je suis daccord, c'est impossible en temps réel, ca a rien a foutre sur un site de jeux, et j'ai deja vue bien plus impressionant que ca (sorry i'm french and i'm too tired to translate!)
Posted on June 28, 2007
Good ..
Hey You can view my Blog its also Good
Posted on July 13, 2007
Holy shit, it's awesome!
Posted on July 20, 2007
Some people need to understand the PHYSICS going on here... Yeah, all kinds of amazing things can be done real time. But those are things that are designed to look as good as possible, while using an extremely simplified (or altogether different) internal mechanism. And as good as they get...and when you watch you can see that they are not as good as the real deal...
I've taken a couple graduate computational physics courses, so here are some real benchmarks for you.
200,000 particle simulation with only gravitational interactions (much simpler than anything that goes on in a fluid or flame simulation): 2-4 hours on a generic P4 processor. And that was just a baby thing that looked like crap. Full simulations are indeed usually ran on super computers. (But of course, that point is almost moot...200,000 particles wouldn't be nearly enough, but fluid dynamics isn't simulated using individual particles anyways)
30 seconds of fluid flow in a grid based solver using roughly a 1-10 thousand grids (can't remember off hand) ~1 hour, same processor as before. And again, its just a baby simulation. Nothing even close to the resolution you saw in any of these. And it was in a closed container. No waves. No surface tensions. No splashes and froth. Just raw flows/pressures/densities.
There are better codes out there than the ones I've been using. But not better enough to make order of magnitude differences. Serious physics simulations (for research purposes of course) are generally done on super computers. The videos here certainly don't need that kind of physics detail, and I'd bet were created on a general PC. But they most certainly need more than what I mentioned above, and so were not done real time. The above times were just for physics data creation, no graphics whatsoever involved.
Posted on August 09, 2007
Wow Love The Blog Comment from: Blackpool Hotels I must say remember seeing some fluids related stuff at nvidias site showing some apparently realtime fluids. However I don't know what system they used (or how many of their graphics cards they linked together to produce these results. Thanks Hotels Blackpool
Posted on September 11, 2007
To clear things up. All of these videos were simulated liquid physics (or liquid dynmaics). The 3rd one was from blender, but the others were from a high end program called Realflow 4.
Simulations are done in real time, BUT they are extreamly slow and CPU taxing. Its a VERY intense CPU process, not much so for RAM and video card. To get realtime results with fast performance and such we would need much more advanced and powerful CPUs than we have right now. So ya, in like 10 or so years this might be possible in games.
As for now... no an Agiea physx card woudnt do it, its for AGEIA physX supported games, few of..and is meant mainly for solid object and cloth physics. There wont be realistic looking real time water physics for a while.
btw u can check my site, ive done a few liquid simulations. www.freewebs.com/jsstudi0s
Posted on September 15, 2007
Hello all!
Posted on October 27, 2007
Post a Comment