2013 vs 2012 speed comparisons

Discussions about the use of Realflow
MikeZ
Posts: 21
Joined: Sun Mar 07, 2010 8:56 pm
Contact:

2013 vs 2012 speed comparisons

Postby MikeZ » Fri May 31, 2013 3:51 pm

https://vimeo.com/67400968

My coworker and I set up some realflow simulations to compare 2012 to 2013 hybrido simulations. We tried to keep the same scene scale, cell size, particle count, and object properties the same between the two programs. Since 2012 doesn't have cell size for objects, I assume it uses the domain cell size for collision so we matched the object cell size to the domain cell size in 2013. For the most part 2012 was always faster, although I feel 2013 is more stable and looks better. 2013 did complete the Monkey simulation faster, which I attributed to the adaptive grid size since 2012 has to calculate the full height of the container for the duration. But if that is the case then the ship rising should have also been faster in 2013, which it wasn't.

Can anybody here give anymore information on speed of sims? Am I missing something that will speed up the 2013 sim times?


fkfx
Posts: 309
Joined: Wed Apr 11, 2012 11:11 am
Location: Cologne, Germany
Contact:

Re: 2013 vs 2012 speed comparisons

Postby fkfx » Fri May 31, 2013 5:08 pm

Interesting results, but you are missing something very important with the new workflow, so I think the experiments are not valid.

Due to the nature of FLIP in Hybrido2, it allows you to simulate lots of particles (speaking so I also mean detail) in relatively coarse grids. So instead of increasing the Grid resolution, you will more likely increase the particle count in the emitters. Also make sure you set your object's cell size values no smaller than the grid's cell size value.

MikeZ
Posts: 21
Joined: Sun Mar 07, 2010 8:56 pm
Contact:

Re: 2013 vs 2012 speed comparisons

Postby MikeZ » Fri May 31, 2013 6:21 pm

I thought that the cell size directly related to the collision distance for objects. If I use a more coarse grid and turn up the particle count for the detail, I might get a faster sim, but won't I have a less accurate simulation because the collision distance is larger?

MikeZ
Posts: 21
Joined: Sun Mar 07, 2010 8:56 pm
Contact:

Re: 2013 vs 2012 speed comparisons

Postby MikeZ » Fri May 31, 2013 10:06 pm

https://vimeo.com/67429689

I resubmitted my Monkey simulation using your suggestion. I increased the cell size from .03 to .06 and then increased the total particles from 10.2 million to 14.4 million.

The total sim time was 1h 32m 16s, more than an hour faster than it was before, but there is a lot less detail. There is less of a splash when the monkey first hits the water, and when his tail skims the water later, there is almost nothing, whereas before there was a nice splash.

cgnema
Posts: 9
Joined: Tue May 28, 2013 10:35 pm

Re: 2013 vs 2012 speed comparisons

Postby cgnema » Fri May 31, 2013 10:09 pm

It's the secondary emitters that are MUCH faster than in 2013. Cache those domains and try simulating a splash in 2012 and 2013, I'm curious how that will go.

User avatar
Oldcode
Posts: 716
Joined: Thu Nov 03, 2011 11:28 pm

Re: 2013 vs 2012 speed comparisons

Postby Oldcode » Sat Jun 01, 2013 12:39 am

Any difference with 2013 for standard particles?
"My sword and my service are yours."

"Then forever remind him of his vow, Knight of the Old Code."

fkfx
Posts: 309
Joined: Wed Apr 11, 2012 11:11 am
Location: Cologne, Germany
Contact:

Re: 2013 vs 2012 speed comparisons

Postby fkfx » Sat Jun 01, 2013 5:51 pm

Bounce up the surface particle sampling even more and make the object interaction factor for the tail stronger. Besides, 15 Million particles is an easy job for Hyb2...Try 100 Million particles and more and then you'll there is a big advantage over Hyb1.

@Oldcode:

Nope, AFAIK there weren't any changes made to the SPH in this release, apart from the very powerful Scripring Nodes.


Return to “Realflow”

Who is online

Users browsing this forum: Google [Bot] and 1 guest