nVidia's realtime fluid physics

[thilanliyan]

www.rockpapershotgun.com/2013/04/25/making-a-splash-nvidias-real-time-fluid-physics/

The water (assuming it is supposed to be water) definitely looks better than Cryostasis for example, but it still seems a bit too viscous...still, good stuff.

 

[boxleitnerb]

These demos are always nice, but we never see that tech in games

 

[bystander36]

These demos are always nice, but we never see that tech in games

That's not entirely true. It just takes several years before they start to get added. In their demos, they can focus entirely on one effect, in a game, it takes too much power from other things that are going on. As hardware and tech improves, we start to see it a few years later.

 

[SirPauly]

I think the frame-rate may suffer on current hardware but something that may be viable a generation or two from now.

Here are some more videos of Position Based Fluids:

http://www.youtube.com/watch?v=F5KuP6qEuew&feature=player_embedded

http://www.youtube.com/watch?feature=player_embedded&v=mgYztcjOvRQ

It also looks like they may be an APEX module for hair/fur as well.

 

[3DVagabond]

I'd seen this demo reported on another site (don't remember which right now). The strange thing is that they are using a GTX-580, not a Kepler card for the sim.

 

[tviceman]

Best looking simulation of water I've seen yet, but yeah it will take a few years before that can run at 60fps in a game with other stuff happening on screen.

 

[Arkadrel]

What makes this water demo differnt from all the other water demos,
we've seen? Visually it doesnt strike me as impressive (seen other water ones looking as good).

What from a technical standpoint makes this one differnt? or worthy of notice?

 

[thilanliyan]

I'd seen this demo reported on another site (don't remember which right now). The strange thing is that they are using a GTX-580, not a Kepler card for the sim.

Isn't the 580 faster at these type of computational workloads than Kepler, bar Titan? Maybe that's why.

 

[3DVagabond]

Isn't the 580 faster at these type of computational workloads than Kepler, bar Titan? Maybe that's why.

That was my first thought, as well. Don't know for sure though.

 

[Imouto]

http://www.youtube.com/watch?v=U3acQ5dDKEs
http://www.youtube.com/watch?v=-IiRzmfs5aw

Nvidia sure knows how to turn you on.

 

[bystander36]

If that is CUDA, then I'm not so sure it is faster on a 580, as every core on either card can due CUDA. If it is OpenCL, then the 580 would definitely be faster.

 

[tviceman]

Isn't the 580 faster at these type of computational workloads than Kepler, bar Titan? Maybe that's why.

Why does this keep getting perpetuated? Show me one game that the gtx580 is faster in. More specifically, show me one game that uses GPU physx and is faster on gtx580 than gtx680. I will save you the trouble and tell you there isn't any. GPU physx is heavily reliant on shader power and gtx680 scales normally on performance in gpu physx with respect to gtx580.

 

[3DVagabond]

Why does this keep getting perpetuated? Show me one game that the gtx580 is faster in. More specifically, show me one game that uses GPU physx and is faster on gtx580 than gtx680. I will save you the trouble and tell you there isn't any. GPU physx is heavily reliant on shader power and gtx680 scales normally on performance in gpu physx with respect to gtx580.

He wasn't talking about games. There are compute functions that the 580 is faster in. This might be one of them, is all that's being considered.

Nobody is perpetuating anything. It's a curiosity as to why they would have used a 580 and that's just a possibility. Do you know any reason they would have used a 580? If so, that's what's being considered here.

 

[SirPauly]

What makes this water demo differnt from all the other water demos,
we've seen? Visually it doesnt strike me as impressive (seen other water ones looking as good).

What from a technical standpoint makes this one differnt? or worthy of notice?

http://mmacklin.com/pbf_sig_preprint.pdf

 

[Will Robinson]

He wasn't talking about games. There are compute functions that the 580 is faster in. This might be one of them, is all that's being considered.

Nobody is perpetuating anything. It's a curiosity as to why they would have used a 580 and that's just a possibility. Do you know any reason they would have used a 580? If so, that's what's being considered here.

The Fermi series card's weak compute performance is a bit of a touchy subject it seems....:\

 

[3DVagabond]

The Fermi series card's weak compute performance is a bit of a touchy subject it seems....:\

I think you mean Kepler?

 

[bystander36]

Since these are PhysX demos, I don't think Kepler's compute performance matters. PhysX uses the normal CUDA cores as far as I'm aware.

 

[Ibra]

http://www.youtube.com/watch?v=U3acQ5dDKEs
http://www.youtube.com/watch?v=-IiRzmfs5aw

Nvidia sure knows how to turn you on.

Lol, noob. Takes movies rendering programs and compare with Pysx SDK.

Best Game of the Year 2013, Best Graphics winner is Pixar Studios.

 

[Imouto]

Lol, noob. Takes movies rendering programs and compare with Pysx SDK.

Best Game of the Year 2013, Best Graphics winner is Pixar Studios.

Bullet Physics, built-in Blender physics engine.

This tech demo is nothing special. In fact it's a lame attempt to stay relevant.

 

[Pottuvoi]

Very nice looking demo and tech should be implemented in physx within a year.
Thus if games like borderland2 used the previous version, we should be seeing the improved one as well.

additional videos.
http://www.youtube.com/watch?feature=player_embedded&v=UDqtEFDCSlc
http://www.youtube.com/watch?feature=player_embedded&v=LhcgjPFykXY

 

[Granseth]

I really hope nVidia's PhysX (and other solutions) find a way to make water look and behave better while not being too taxing. And I love that they are researching variants to make that happen.
And even though I'm not a big fan in proprietary solutions they will hopefully put pressure on the open solutions to make better water simulations.

 

[3DVagabond]

http://www.youtube.com/watch?v=U3acQ5dDKEs
http://www.youtube.com/watch?v=-IiRzmfs5aw

Nvidia sure knows how to turn you on.

Both of these renders are way better than the nVidia demo. Difference is, and it's a huge difference, these weren't done in real time. If you look at the second link, at the end of the playback it states file sizes (Fluid cache was 6.4GB and the scene file was 57.2GB) and render time of 220+ hours! That was for ~31seconds of playback. Let me repeat that, 57GB file size, 6.4GB cache, 220+ hours to render ~31sec playback.

 

[Imouto]

Both of these renders are way better than the nVidia demo. Difference is, and it's a huge difference, these weren't done in real time. If you look at the second link, at the end of the playback it states file sizes (Fluid cache was 6.4GB and the scene file was 57.2GB) and render time of 220+ hours! That was for ~31seconds of playback. Let me repeat that, 57GB file size, 6.4GB cache, 220+ hours to render ~31sec playback.

If you think that rendering realtime is rendering a bucket of water at 15 fps with a GTX 580 it's cool.

Again, just rendering a bucket of water, nothing else.

 

[SirPauly]

Nvidia sure knows how to turn you on.

Absolutely! In real time!

 

[f1sherman]

If you think that rendering realtime is rendering a bucket of water at 15 fps with a GTX 580 it's cool.

yeah well that's still 15 frames per second compared to your Bullet/Blender:

"One frame out of 710 took about 22 minutes to render"

Quite a progress I would say