Maya Particle Doesnt Support Multi Processor Or What

[12monkey]

Iam using maya particle to simulate sand, but maya only use 10~15% of cpu resource.

I have 2 Xeon E5410 cpu, 4G FBD ram, using maya 2008

Anybody know why? And how to fix it?

[12monkey]

This 300 frames simulation took about 5 hours to finish, if maya particle could use all cpu it could be much faster

[deke-kincaid]

Currently Particles aren't multithreaded. There is no way to fix this. Actually up until a couple years ago only rendering was threaded. They are slowly one by one threading all the other operations in Maya.

[12monkey]

Thanks for the answer beaker! Then I really hope particle will support multithreaded in next release. It really feels like the more I use other 3D software the more I feel maya is not up to date.

[deke-kincaid]

QUOTE(12Monkey @ 06/13/08, 07:41 PM) [snapback]287053[/snapback]
It really feels like the more I use other 3D software the more I feel maya is not up to date.
Actually Maya is more multi threaded then most other apps. Most applications only threaded at the renderer level. Maya just started threading other major features in 8.0.

I actually don't know any applications that thread particles at the moment, not even king of the particles Houdini. Simulations are especially hard to thread because you have to figure out one frame before figuring out the next so you can't just give a frame to each core/proc.

[joojaa]

QUOTE(beaker @ 06/16/08, 01:31 AM) [snapback]287103[/snapback]
I actually don't know any applications that thread particles at the moment, not even king of the particles Houdini. Simulations are especially hard to thread because you have to figure out one frame before figuring out the next so you can't just give a frame to each core/proc.

I do, however it depends heavily on how the particle dynamics are done and what they do. If they are pure expressions dirving/forces them you could rather easily multitherad them, even within maya. (what you do is you give each register slice to one proc, problem arises whan you want things to interact in the second degree)

But again if you want collisions/interaction etc it becomes tricky. I work for a company that basically ONLY does simulations (simulation is a wide thing) and unfortunately theres no easy way to make a discrete simulation fully threading. Some areas can be made easily but not all, and unfortunately flexibility counts more than threading for most users.

Anwyay quite often its not just the speed of the individual particles thets the problem but rether the environment in how they are executed.

This issue is by the way one reason why theres 2 types of supercomputers, massive parallel processors and massively fast individual processors. And unfortunately not everything is up for parallelization.

there are other areas too such as video encoding taht quite often does not thread either.

PS: 5 hours isnt all that much, we have clients that do simulations that literary take 1 month to do, on the best money can buy.

PPS: quite literary having more procs needs to change the way you work, because you need to be blasting things in the background more. So work on something eslse while you render the particle caches.

[12monkey]

Thanks beaker and Joojaa for the answer!

I dont know much about threading, but thing I do know(or I saw) is when I use max particle I can jump to any frame I want instantly, and I can apply force to particle at any frame and see the result instantly too. Or may be max particle is not particle?? And what about fluid? Fluid have to figure out one frame before figuring out the next, so why fluid support multi procs?

[joojaa]

Ahh so you jsut want runup to be canceled

QUOTE(12Monkey @ 06/17/08, 11:00 AM) [snapback]287152[/snapback]
I dont know much about threading, but thing I do know(or I saw) is when I use max particle I can jump to any frame I want instantly, and I can apply force to particle at any frame and see the result instantly too. Or may be max particle is not particle??

You can do that in maya too, just reduce it all into a determined expression! See the problem arises when you make something into a feedback loop, that depends on previous frame. This will cascade from frame to frame. This issue has NOTHING to do with threading or multi processing.

Aslo you would want run up to be disabled, off course just because it updates does not eman its the right solution.

So example:

CODE
file -f -new;
particle -jbp -0.0 0 0.0 -nj 1000 -jr 1 -c 1;
dynExpression -s "particleShape1.position=particleShape1.position0+<>;" -rad particleShape1;
dynExpression -s "particleShape1.position=particleShape1.position0+<>;" -c particleShape1;

is instantaneous, because all calculations can be done on the frame. Maya could be implemented to update things like gravity in similliar fashion. The problem arises when you mix 2 forces suich as turbulence in the flow, its now not totaly clear where the particle will end up, so theres no other option than to step trough all frames.

Now let me clarify this in a expression version:

CODE
file -f -new;
particle -jbp -0.0 0 0.0 -nj 1000 -jr 1 -c 1;
dynExpression -s "particleShape1.position=particleShape1.position+<>;" -rad particleShape1;
dynExpression -s "particleShape1.position=particleShape1.position+<>;" -c particleShape1;

this works differently if you jump in time than it does when played trough every frame (yes it even produces different results for different frame rates at so and so seconds!). This simulation can be said to have a side effect, of needing to run all frames.

Tip: When you run simulations of huge amounts of particles up to a cache disable your ogl viewports as they will eat your time.

PS: So maya users pay for the flexibility they gain as truly NO edge of this can be gained without understanding the problem.

[12monkey]

Thanks Joojaa for the answer!

But the two example you provided are the same, you sure you not miss typing anything?

You side the problem arises when you mix 2 forces such as turbulence. But in max I can attach not just turbulence field, I can attach any number and any kind of field at any frame, and I can play with parameters of any of those field at any frame and see the result almost instantly. eg, 20k particles with 4 or 5 fields at frame 100(particles are emitted from emitter), and I change a parameter of one of those fields may be strengh, then it only take about 1 or 2 secs to update, so iam sure its not a runup. To me, it seems in max every single particle is predefined or expression driven just like the example you provided. It seems like max particle system and maya particle system is totally different thing.

[joojaa]

ohh sorry teh firt one was missing a 0 at the end of its position fixed now.

QUOTE(12Monkey @ 06/17/08, 08:44 PM) [snapback]287170[/snapback]
You side the problem arises when you mix 2 forces such as turbulence. But in max I can attach not just turbulence field, I can attach any number and any kind of field at any frame, and I can play with parameters of any of those field at any frame and see the result almost instantly. eg, 20k particles with 4 or 5 fields at frame 100(particles are emitted from emitter), and I change a parameter of one of those fields may be strengh, then it only take about 1 or 2 secs to update, so iam sure its not a runup.

But yes mayas execution freedom has a slight disadvantage on massive update. Maya was never designed to do things like this in long pipes. And yes the particle system is starting to be quite old by now. It however kicks maxes systems ass for most things any day if not in execution speed but rather freedom of choice.

The problem isnt getting instant feedback, thats easy. but getting the 100% right instant feedback is another thing entirely just cant be done very easily. Maya authors just chose not to make you feel good for getting feedback and wanted you to see the drawback too.*

See even maya could cache the last time step and guesstimate it on changed values, however to run the right piece of information to frame 100 would still mean stepping trough all 100 frames.

QUOTE
To me, it seems in max every single particle is predefined or expression driven just like the example you provided. It seems like max particle system and maya particle system is totally different thing.

Well if thats the case then you lose about 99% of the entire idea of using particles in first place. Either things are a discrete simulations or they are not. If not then you loose everything you gain by Mayas particle system.

however this doesnt mean max doesnt run it up, i suspect it does (or i know). Maya just doesnt bother to do anything untill the user tells to do so, after all emitting 5000 particles to frame 100 with 2 turbulences and gravity takes about 1-0.5 second on my computer. Maya just choses not to do taxing calculations unless one must. So maya just waits for you to update its state.

It s no big thing to let maya do this automatically. Maya still needs to step those frames tough.

PS: multit threading wouldn't make the 5 hours much faster you'd still be in a 1 hour time frame even on monster 8 proc system. Thats the problem a machine form 2003 isnt significantly slower than a mchine of 2008 atleast not when compared to the machine in 1998. Because therading does not upscale the way we would like.

*this seems to be the difference between maya and max. In max the maximization is on how fuzzy you the user get when using the software. I don't dispute it doesn't make for a wonderful experience. Here maya and houdini have things in common, they dont really care of the users feelings what they care of is how things work. So at the end of the day it is easier to cope with a problem in maya and houdini than in max, because atleast you can work with how things work.

[12monkey]

Thanks for the answer Joojaa!

QUOTE
Well if thats the case then you lose about 99% of the entire idea of using particles in first place. Either things are a discrete simulations or they are not. If not then you loose everything you gain by Mayas particle system.

Yes I agree with that. But what I said is only based on what I saw and guess, since I just started to use max, and not just max, I also tried XSI and houdini. Because to get something done could be far too easier in one software than in the other. eg, fumeFx in max is far too user friendlly and much much faster than maya's fluid, its like 2 mins with fumeFx and hours with maya fluid(from a artist point of view).

QUOTE
Here maya and houdini have things in common, they dont really care of the users feelings what they care of is how things work. So at the end of the day it is easier to cope with a problem in maya and houdini than in max, because atleast you can work with how things work.

And yes, I totally agree with that.

And BTW, I got some problem in this pose too:
http://www.highend3d.com/boards/index.php?showtopic=245291
So if you got the time... it seems no one bother to reply that.

[joojaa]

QUOTE(12Monkey @ 06/18/08, 04:56 PM) [snapback]287225[/snapback]
And BTW, I got some problem in this pose too:
http://www.highend3d.com/boards/index.php?showtopic=245291
So if you got the time... it seems no one bother to reply that.

it was so fantastically overloaded that i got headache reading it so i decided to skip it... might look later.