Varying simulations from same setup

[pbuckley]

I have some strange things going on here - I'm hoping somebody can shed some light on this.

basically i set off a couple of big simulations off overnight - long sims with small spacing settings - big slow sims!!

when i came in the the morning deadline had reset the task to begin again - (an 8 hour time out that i wasn't aware of - now fixed)

but the issue is - the new sim which is simming and the original sim that got stopped don't match.

so say my original sim got to F200 and my new sim (going overe the top of the fxd files) is only up to F156 - then there is a big change between F155 and F156 when I render.

I was under the impression that you could sim the same file ahundred times and get the same result (so long as you don't change any settings of course!!)

Another strangeness is - another sim that was also huge didn't display this behaviour - i.e. the new resimmed frames matched the older simmed frames perfectly - as i would expect.

(we also have done some tests on smaller scenes just to try and work out the issue - and we have the same results - it seems random, sometimes the sim is the same and sometimes not.)

anybody got any ideas what may be going on here?
p.s thanks to anybody for reading all of this!

[Rif]

on testing phases .. i remarked this behavior on Big Setups.. Because it was hard to reproduce it ( not everyone can sim those big 4 gig ram ++ setups) , i think the "bug" failed to be adressed .. if you can , by any chance , reproduce this behavior on a simpler\smaller setup , it would worth a while to send it directly to mailtoupport@afterworks.com or\and mailto:Cgfluids@cgfluids.com .. For now i would highly suggest to not rely on sim resuming when ram usage passing 4 gig + ..

Also , are you using the Resume Sim button ? or the Start from selected fxd file option ? i think you have more chance to Continuis simulation with no pop with the resume button..

my 2 cents

[pbuckley]

thanks for the response rif.

the restarting simulations was not stopped by myself and not rekicked off by myself - purely deadline quitting a task due to a timeout and then restarting it, so there are options of starting from a preset .fxd file or resuming.

problem is we are getting this weird behaviour even with simple files. a co-worker here set off a very simple sim - and rendered it. - then set the sim off again (after deleting the original fxd files) and the two simulations are different!!!

I could send this simple file to the afterworks people but it is very simple - no special setups, nothing kooky going on.

are we going mad?!

[Rif]

"problem is we are getting this weird behaviour even with simple files. a co-worker here set off a very simple sim - and rendered it. - then set the sim off again (after deleting the original fxd files) and the two simulations are different!!!"

hold on a second.. This is Normal from what i understand .. 2 simulations will not be 100 % the same exact simulation in a Voxel Based simulation .. i think that might be due to simply the of the Xyz noise turbulance and \ or the Per Source Turbulances , and or the Burn rate Variation Seeding differently.. i mean it should be the same Setup look wize and timing , but Comparing File render A and File render B of 2 re-run of the same Scene should not be expected to match .. i though you where talking about something else ( more stoping a big sim and Resuming it or continuing it and suddently in 1 frame poping out..

[pbuckley]

Hi rif.

that seems really strange to me - surely the same values, turbulences,sources etc should produce the same results? It is surely still reproducible maths that calculate the forces/heat etc etc- maybe I'm just naive!!
problem is that it seems a little inconsitant and unpredictable, as sometime the solves have been exactly the same, and sometimes not.
just trying to get a handle on what may be the trigger for these differing outcomes.

thanks again for the response rif - all insights are greatly appreciated.

[sitnisati]

Hello Paul,

Can you please email me a simple scene with steps to reproduce it ?
I'll take a look here and think of what migh happnening.
How many cores do you use ?

Thanks

Kresimir

[pbuckley]

hi Kresimir,

I have already sent a couple of emails with various quicktimes and files to support.
we also sent some screen captures to show the results we were getting.
i have had some correspondance with Vjekoslav Brkic and am waiting on a response from the last stuff we sent through.
i had Grant Adam -our main TD guy here run through a pile of tests - he was experiencing the same things.
at the moment we use 8 cores in the sim - but if you have a chance check the last email to support grant was finding difffering results depending on
changing the affinity via the task manager.

thanks Kresimir.

PB.

[sitnisati]

Hi Paul,

I've got files and was able to repro it.
We're looking into it.

Thank you.

Kresimir

[bdemetz]

Just to 2nd this, but on a different note:

2 different machines seem to sim the same sim differently even though the same machine will resim exactly the same. I've run the same sim (that uses around 7GB of ram) on the following 2 machines and the sims look different):

--dual quadcore 3ghz, 8GB of ram
--dual quadcore 2ghz, 16GB of ram

Both sims look decent, but I was also hoping to get the same result on both - even the shaders look slightly different.

Thx again,

[Rif]

ya it seems to be the MultiCore differences .. But i do not and never did really care of this.. As far as i know , maya does that too right ?

houdini dosnt assure with their grid based solvers the same sim either i think .. i do not personnally see an urgence their ..

[bdemetz]

true true.

i guess in my particular case i was hoping to resim an explosion that was just perfect, and when i saw the change in shape and the fire shader looking different - i couldn't let go of the original ; ) even though to most people it probably looks fine.

i guess the rule is - if you sim and u love it, either save the cache or render it NOW - you may never get it again.

[pbuckley]

rif - I can't understand why you have no problem with not getting predictable repeatable results with the same file.
If you have a sim which is perfect and for any reason you lose your fxd files you're hosed.
the director comes in and says I want that explosion exactly that you guys did 6 months ago for me - and we say sorry no can do.
we have all the the scene files but they probably won't give you what you got last time.

it makes it very hard to judge complex long sims against each other - does that look different due to the advection stride and fuel and temperature differences or is it just different because it is - tough.

[Rif]

Maybe thats the reason :

I never think Anything we do is perfect ..,.

So , as the re-simming is 90 % similar , i m fine on improving from my old setups .. even if i cannot back up my older sims ..

but thats just me..

cheers,

[Rif]

Of course though.. if it Can be repeatable , then sure .. it was , long time ago , in a different Solver under Alpha Phases if i recall properly .. but it was also less MT than what it is now .. i rather have Sim power than Reproducability power ..

[pbuckley]

Can we have both?!
yeah I know. i'm greedy!

[bdemetz]

Even without MT, I'd still love to have the old solver as an option. I still have a lot of scenes that used it. It had a certain look that was not obtainable with CG.