Why are dynamics in Houdini slower than other softwares?

[rohandalvi]

hi,

Firstly, just to make things clear, I really like houdini and i dont really want to start a war, just looking for opinions.

I recently purchased the Digital Tutors DVD - Intro to Dynamic in Houdini.

It's really cool. But I noticed something, In cases like cloth and wire solvers ( and not to mention fluids, but it's only the first version so it's ok) Houdini is a lot slower in comparison to other softwares.

For example, In cinema4d I was able to to collisions with grass and a sphere in realtime, without any problems, even on my older machine, the same in houdini through wire solvers takes a lot longer to calculate.

Also, cloth in comparison to something like NCLOTH from maya is very slow.

My, question would be, For a software which is known mainly for its particles and dynamics, shouldn't the dynamics be a lot faster?

Just wondering.

Any opinions from people actually working on houdini professionally , since i mainly use it as a hobby, would be most appreciated.

with regards

Rohan Dalvi

[rob]

I think you should post up some test scenes with your dynamics set up then people can test them on various OS's.

R

[rohandalvi]

hi,

I couldnt really post any complex scenes, it's mostly just the tuts from the Digtial tutors DVD. I started doing dynamics 3 days ago.

I have no problems with houdini spending time to calculate the simulations, they are very accurate, but it just seemed slower when you compare it to someting like Ncloth in maya which is gives you the result in real time.

the biggest one i noticed was the wire solver. I am uploading a simple scene. let me know if I am doing something wrong.

but it's just that cinema4d hair seemed faster, well not faster , it did the calculation realtime. houdini took about 2-3 mins to do the calculation. I know thats not like a lot of time, but still seems slower than realtime.

I dont have a fast machine, it's an AMD 2800 with 1 GB of ram.

I think what I am saying is that sidefx should boost up the speed of Dynamics as the sims in maya especially stuff like Ncloth can now do stuff almost realtime.

with regards

Rohan Dalvi

grass.hipnc

[jason_slab]

yes Houdini is slower, but what i find and i can't compare this cinema4d, you have alot more control over what you want to do, meaning you can access almost any data in the sim a modify it. i'd love to have the speed of the physX engine with the same level of control that Houdini allows!!

as for cloth, well i haven't tried 9.5, but as far as the tests we did in houdini 9, cloth just isn't usable in a production, unless of course you have 2 years to complete one shot

wires are really cool in Houdini, I've don't quite a few tests using wires as guides for fur/hair, yes it's slower, but it's way more accurate than my tests with shave and haircut(XSI, 3dsmax).

hope that helps

jason

[sibarrick]

AFAIK Sesi is aware that the cloth solver isn't the best and I think they will plan to improve it. They have first decided to concentrate on RBD and fluids which I guess gets more use. I'm sure they will return to cloth in H10 and make improvements.

I wonder too whether DOPs is like mantra it can appear slower on small test scenes but scales up with less loss of speed.

Your simple test with a sphere might run very quick in cinema4d but give it 10,000 spheres and see if it chokes....

Don't forget too there are a lot more options to fine tune things in Houdini, just changing the way collision geometry is built will make huge differences.

[rohandalvi]

thanks a lot guys,

I guess those are the answers I was looking for.

I do agree that houdini is way more accurate, doing hair collisions in MAX is plane horrible. they just go all over the place.

I am very new to dynamics in Houdini, so i am only starting to explore them. But they are really easy to setup due to the shelfs and are really accurate.

I just hope they make them faster, it would be really cool.

with regards

Rohan Dalvi

[symek]

Like said before, so much depends on setup. Moreother it's not so easy to prepare this setup. One needs to have good understanding of all DOPs, Sop and Script Solvers, and DOPs data expressions. Than you realy can make it flying

cheers,

sy.

[stevenong]

It's fair to say DOPs is slower compared to the competition. I would like to see some optimization myself.

If accuracy is not too critical, you can always reduce the substeps of the simulation. With your file, I reduced the substeps of the wire solver to 1 & the speed seems reasonable, IMHO.

I did some rough timing with my stopwatch & hitting Play so it's not the most accurate.

100 frames as in your file.

Wire Solver Maximum Substeps: 1 --- about 20 seconds

The above is with and without Wire/Wire Collsions (Wire Object > Collisions tab)

Wire Solver Maximum Substeps: 8 --- about 1 min 37 seconds

Again, the above is with and without Wire/Wire Collsions (Wire Object > Collisions tab)

However, with your example, the Wire/Wire Collsions shouldn't matter much. If you're learning & not concerned with matching the simulations in the tutorials, you can always reduce the substeps etc to speed things up a little.

Please post any cool sims you do in the form of a flipbook if you can.

Cheers!

steven