Detail Difference between SPH and PBD?

[Oldcode]

Hi,

I'll be buying 2015 in the next week or so and I'm really looking forward to the speed increases in Dyverso. I'm curious though, is there any differnece in the detail of the fluids between SPH and PDB. Since PDB is so much faster, I thought that might not be as good for simulations that require a fair amount of detail, such as a person being hit with a bucket of water.

What's the difference between the two?

Thanks,

[DeltaJuliet]

Hi, iirc from one of the webinars the legacy fluid type is being removed completely.

This is from RF2015 Manual -

"Liquid - PDB" (Position Based Dynamics) method is completely different from RealFlow's other solvers, it is faster than the SPH methods, supports GPUs, and it is open for other fluid types, e.g. granular substances

"Liquid - SPH" (Smoothed Particle Hydrodynamics) method is similar to RealFlow's standard particles, but Dyverso's SPH is much faster and supports GPUs

Hopefully Luis Miguel will give a better answer as i can't remember where i heard it.

[tsn]

With SPH you'll most probably get more detailed results, but this depends on so many other things. I'd say you can get the same level of detail with PBD as with SPH with a sufficient amount of particles. Anyway, the PBD solver reacts completely different and it may take more test simulations to achieve similar results. With standard settings SPH has a more fluid-like appearance than PBD. With PDB internal and external pressure can help to achieve more watery results, also surface tension. But as in many cases this is also a matter of personal preferences and maybe you have a complete different impression from PBD fluids.

[Oldcode]

Thank you both.

Right now, I'm working on a project where a human figure is coming into contact with a very thick fluid. In RF2013, I'm using standard particles with a viscosity of 200, and a high level of sticky and surface friction to really make the fluid stick to the figure. At the scale I have now, and a distance off set of .01, I get pretty good cohesion of the particles to the figure mesh and a high level of detail in the fluid as it reacts to the figure's movement.

I guess my question is, do you think I can get that same level of detail with the PDB? I only ask because Dyverso PDB is supposed to be faster than Dyverso SPH. Although I'm not a professional, and don't have tight deadlines, going faster is always preferable. As I type this, I'm running a simulation of about 660,000 particles with the viscosity of 200, and the appropriate sub-steps. It's taking between 6-7 minutes per frame to simulate on average. Some frames take a lot longer than that. If I could use PDB, conceivably, the same frame might only take 1.5 minutes. Any times savings would really help.

Oh, with 2015 and high viscosity, how much does high sup-steps slow down the simulation?

Thanks again,

[tsn]

If you work with high viscosity settings I recommend the PBD solver. 660 K particles and viscosity set to 200 shouldn't be a problem at all for Dyverso. Here, simulation speed mainly depends on the complexity of your object, (and of course your computer). Since I don't the specs of your machine it's hard to predict simulation times, but with an i7 you'll be probably under 1.5 min/f.

And be careful with increasing substeps. The PBD solver works completely different than SPH, so higher substeps do not necessarily mean that you'll get more accuracy or better results. In fact you might even lose stability with high settings. You should increase the solver's iterations instead. Anyway, with a viscosity value of 200 you might not see any problems at all, but as said this depends on the entire environment (daemons, objects, etc.). I ran (basic) tests with viscosity values of up to 3000 and didn't encounter any popping particles with the solver's defaults.

[Oldcode]

...Anyway, with a viscosity value of 200 you might not see any problems at all, but as said this depends on the entire environment (daemons, objects, etc.). I ran (basic) tests with viscosity values of up to 3000 and didn't encounter any popping particles with the solver's defaults.

Hey tsn,

Thanks for the info! That sounds incredible! Will the increase in iterations slow down a simulation?

As for me, I’ve got an AMD FX 9500 I think, 4.2 Ghz with 8 cores. My geometry is fairly complex. It a poser figure that I brought into Lightwave, then into Real Flow 2013. Because the polygon count is so high, I have to increase the scale to 25 for the particles to interact with it properly.

Any increase in speed will be very welcome. If instead of 6-8 minutes per frame, I can get that down to 1 or 2 minutes, that would be fantastic! I’ll definitely try out the PBD first.

Thanks!

[tsn]

Your welcome Oldcode.

Yes, higher iterations will increase simulation times, but perhaps you'll succeed with the default settings already. Iterations are used to stabilize the interactions between the particles. It's a bit like with cloth solvers: there you have a constraint solver and this solver requires a certain amount of iterations to create stable result.The more iterations, the slower the simulation.

Scale, on the other hand, is always a bit tricky and you might see some problems, but it's hard to say from a distance and without having it tested myself. Since you're a long-time RF user you'll certainly know how a simulation's environment can influence your results

[Oldcode]

Thanks again tsn,

Here's a picture of the simulation I'm running now. As I said before, it averages 6-7 minute per frame. I wonder what PBD will do!

[tsn]

Thanks for the screenshot

I'm also curious how Dyverso will be performing here.