Different motion with partitions

hi!

I'm new here...  Sorry if it's a stupid question...   But my partitions did not have the same motion than my original particles...  The first frame is fine but after that, the particle move in a different way.

Any idea why this is hapening?

Thank You!

Try incrementing only the position seeds. See if that fixes it.


  • Chad

No change....

I really don't understand...

No change…

I really don’t understand…





Hmmm.

What does the PFlow contain?



Try this:



Open PFlow, go to an operator you think should be randomized between partitions and change its Random Seed by 1 (for example, if it was 12345, set it to 12346).

Save to a PRT file WITHOUT using the partitioning.

Load the original sequence and the manually created partition into a PRT loader and compare. Do you get the same wrong result?



If yes, then there might be something in the logic of your flow that breaks the idea of partitioning. I haven’t seen it before, but one can create rather crazy flows after all.



Also note that you can exclude operators or whole events by renaming them to contain the string “NOSEED” in their name. If you feel just a single operator has to be randomized, you might want to exclude all other events and operators to get what you want.



But without knowing what is in the flow, it is rather difficult to tell why you are getting what you are getting (not that we have seen what you are getting, mind you ;o)



Cheers,

Bobo

No change....

I really don't understand...

I email you my flow…  I will test your recommandation later…

Can you describe a little bit about how your flow is set up?



In my experience, the major reason for choosing when to increment an operator’s seed or not is whether it has a macro-effect, or a micro-effect. As an example, a flow which first creates 200 random particles, then starts spawning a stream of particles from each of those seed particles, is the kind of flow which requires special treatment. In this example, the initial position operator should NOT get seed increments, but the spawn operator SHOULD get seed increments.



An analysis along these lines, determining which randomized operators produce macro versus micro effects can assist in figuring out how to set up partitioning.



-Mark

Another example, I’m working on a scene with Keep Apart. Partitioning doesn’t work there because each partition will apply forces differently because the particles making the force are in different places.



Instead, I use a proxy particle system for the Keep Apart that is set to NOSEED and is set to Cache Disk. This way the forces applied are invariant.


  • Chad

I email you my flow…  I will test your recommandation later…

hi!

I’m new here… Sorry if

it’s a stupid

question… But my

partitions did not have the

same motion than my original

particles… The first

frame is fine but after

that, the particle move in a

different way.

Any idea why this is hapening?

Thank You!





Hi!



I loaded and tested the scene you sent us and it DOES work correctly out of the box if partitioned on the same machine.



Did you create the partitions on the network or on the same workstation?



If your network has multiple machines with different CPU stepping, or even worse, different makers (Intel vs. AMD) or types (32 vs. 64 bit), the results would be different.

We discovered that even our Single Dual Core Sun Systems machines produce different random number results than the Dual Dual Core Sun Systems when using certain Max features, and both are using 64 AMD CPUs!



So if you want to get consistent results from partitioning, you have to make sure all partitions are created on identical machines.



If you DID create your partitions on the same machine and are still seeing differences, then we would have to investigate further - I could NOT reproduce the problem on my workstation.



The screenshots show the two partitions on frame 200 - once overlayed using two PRT Loaders with Use Node Transform off and once side-by-side using the Node Transforms.
The sequences are similar, not identical, which means that the seeding works as advertised when using the same CPU type.

I can't upload file from the job...   Security team block everything...

 

I forward to you a mail I have send to 'Mark Wiebe' with screen grab.  

I made my partitions localy on my machine..  The same one...

I think I will uninstall and reinstall  Krakatoa..

I can’t upload

file from the

job… Securi

ty team block everything…



I forward to you a mail I have

send to ‘Mark Wiebe’ with

screen grab.

I made my

partitions localy

on my machine… The

same one…

I think I will uninstall and

reinstall Krakatoa…





I saw the screengrab and I can see how your second partition is completely different. I just cannot reproduce this here - and I tried multiple times. The two systems remain aligned and similar throughout all frames from beginning to the end.

We will have to scratch our heads a bit more.



Were both partitions created in a single run?

(not that it would matter…)

As I said I will re-install Krakatoa and after that we will see...

I don't find what could cause that...  :(