View Full Version : Not enough RAM for Subpatch?

02-12-2008, 12:09 PM
I have an elaborate object I need to do a fly-around shot on for a personal film project. It is an organic mass, kind of Roger Deanish (http://www.rogerdean.com/architecture/index.htm) planet, and it has all kinds of stairways pathways and caverns and so on. So I have begun building it as one organic flowing piece in subpatches.

I am up to a point where I have about 280,000 polygons all one contiguous mesh.

Now when switch into TAB mode, I get a crash in modeler. I can only quess that this is a RAM limitation because lower poly versions of it don't crash.

Any recommendations? I am still 32 bit. Will a 64 bit OS with more RAM be the answer here?


02-12-2008, 01:37 PM
A little update. The object I was trying to subpatch had stuff in other layers as well. So I took the 280,000 polygons, put them into a new object, closed Modeler to make sure that all copy and undo buffers were clear, opened this new object up and I was able to subpatch without a crash. But taking this geometry and mirroring, it making it twice as large, in this new object, did produce a crash. So it does seem to be a RAM issue from what I can tell. So I wonder if anyone has any experience with objects this large?

02-12-2008, 01:47 PM
I've rendered scenes with upwards of half a million or even 700,000 polies before (no problems at all) but I've never tried subpatching something that large.

02-12-2008, 05:47 PM
probably a painful work around, but it may save time in render to use a finalized frozen mesh, rather than the subpatched one.

look forward to some shots, i like his style.

02-12-2008, 06:23 PM
Thanks guys,

Yeah inigo07 me too. It is not the final render that is the problem here though, it is the initial subd object.

Objuan, right that would be painful and in that case I could use a low res proxy object to set up the scene. That is an option, though I know I am going to need more RAM to render it anyway.

I am mainly curious if anyone has any experience subpatching something this big - and bigger actually because the final is going to be over a million unsubpatched polys.

Any 64 bit users out there?

02-12-2008, 07:56 PM
I have Vista 64, 4GB RAM, and Lightwave 9.3.1 64bit/FPrime. Want me to try something specific?

02-12-2008, 08:26 PM
Hey, cool. Well I can't let this object out because I am keeping certain aspects of the concept of this film under wraps. But I am looking to see if it is possible to subpatch say 1 million polygons in one object. If You could array a bunch of segmented cubes or spheres and see how many polygons you can subpatch in one go before modeler crashes that would be a great help.

02-12-2008, 09:39 PM
Created one tessellated sphere, 162 polys

Used the following Arrays:

10x10x10 for 162,000 polys
before: modeler RAM use 213MB
time: 2-3 seconds
after: 411MB
Worked fine

15x15x15 for 546,750 polys
before: modeler RAM use 648MB
time: 12 seconds
after: modeler RAM use 1.2GB
Worked fine

20x20x20 for 1,296,000 polys
before: modeler RAM use 1.4GB
time: 1.2 minutes
after: modeler RAM use 2.9GB
quad-view window refresh issues, application no longer responsive
Windows lightly banging the page file

I waited 5 minutes, but Modeler just kept flipping between the 4 windows, alternately drawing each one, about 15 seconds between. Probably more an OpenGL issue than anything else, but my machine was using 3.8 of the 4GB installed, so it wouldn't have been too useful anyway. I was able to pick New Object from the file menu and it became responsive again, even with Vista basically out of memory.

Anything else I should try?

02-12-2008, 09:51 PM
Wow cool. OK that helps a lot. So 4 gigs is barely enough for 1.3 million polys that are then to be subpatched.

So then what level of subpatch? The default 3 I presume?

Thanks for running those tests. That is a big help.

02-12-2008, 10:03 PM
2 Subpatch divisions, 3 Catmull-Clark Level
SubD Type Subpatch

I can run it again with 3, but I imagine it'll just blow out sooner. :)

02-12-2008, 10:12 PM
Just for fun, I changed Subpatch Divisions to 3, redid the 15x15x15 test, then did a Freeze to see just how many polygons would result. I guess I could have just checked the count with a calculator, but I was curious.

It's been about 10 minutes, Vista's page file is sitting at 6.5GB, RAM is pegged, both cores are flopping like dead fish, and I have no idea how long this is going to take. :)

I bet if I had 8GB (and that's coming soon), this would probably work just fine.

02-12-2008, 10:13 PM
Yeah thats right.

Actually I only need to get 1 out of it and then use layout APS to make detail dynamically so that's looking pretty good. I am going to go for 8 gigs of RAM I think. Hopefully that will be enough.

Thanks again.

EDIT just saw your newer post. Thanks for running the other test.

02-13-2008, 07:59 AM
Richard, this might be a stupid suggestion, but is there anyway you could break that part of the model into a few sections? maybe at crease or seam areas or around molding on a wall, etc. that would not show the break in the model? Even if you could get it to the point of having it subdivide in modeler, would you really want to work on it where every time you hit the tab key you worry that it will *poof* and crash?

02-13-2008, 12:02 PM
Hi jeff,

Yeah, A little explanation. And also to answer your question. Yes, I have thought of that. I could redesign the entire thing to be separate objects with creases. It would be far from ideal because that was not the vision. But it is obviously an option when the rubber hits the road. (but I think layout would crash trying to render the whole thing anyway.)

Now the explanation so you don't get the wrong idea. Actually the thing has symmetrical elements that join 4 ways, then 3 ways and then again 2 ways. So I am able to build basically 1/24th of the model to start. This section is then radial arrayed 4 times and points welded to make one whole. I can build this far and subpatch to check no problem. Then I can do the other "array" 3 ways and weld and join, again just under my RAM OK.

The problem arises when I have to then take this section and "aray" 2 ways basically a mirror and I have doubled the size.

The entire thing is designed to fit togather this way.

So there are stages:

Stage 1: The 1/24 piece

I am almost complete with this stage.

When done I will array this 4 ways and weld.

Stage 2: Then this piece will array 3 ways and connect to make a final larger piece that I will then add geometry to once I am at this stage that will be unique and not be array-able within this piece if that makes sense.

Once this is done the piece is virtually complete. Except for for the final stage.

Stage 3:I then take this final pice which designed to "array" once to make the final whole.

This is the part where I have the problem and considering what I have tested so far makes about 500,000 polys I predict that the final object will be over 1 mil with all the geometry added at stage 2 and then duplicated at stage 3.

OK, so it is not really a work flow issue as I really rarely have to even look at the subpatches since at the poly design stage I already have a pretty good idea what it will look like and certainly at the 2/24 or even 1/6th stage this is not a problem anyway. Then when I get to stage 2, I will of course work in sections and test the subpatches as I go then connect when done.

The problem is actually having the entire mesh connected at the final stage and being able to press tab at level 1 to get a subd object for layout.

And finally since these "errayed" peices actually make up an organic mass that flows in curves when subD'd they cant be actually separate.

Make sense? :)

02-14-2008, 07:21 AM
^yeah, that makes total sense! :)

Like you mentioned before you could set the subdivide level as low as possible in modeler and then use APS in layout for render time. But sounds like you're gonna have one huge poly model in the end either way!!! Can't wait to see what your working on! (I was always a fan of all the YES album (http://images.amazon.com/images/P/B00007LTIB.01._SCLZZZZZZZ_.jpg) covers!)

02-14-2008, 10:59 AM
Yeah that's cool. Me too. I was influenced quite a bit by Dean when I was doing art in school.

And yeah I am looking to upgrade to an Intel Extreme quad core with 8 gigs of RAM and a 64 bit OS. Hopefully that will be enough for some of the bigger scenes I am working on. I am also looking at possibly getting a board that can handle more RAM. Not sure yet.

02-17-2008, 10:24 PM
A little update. A local electronics supperstore is having a sale this weekend. I picked up a great deal on 8 gigs og 800mhz ram for $172.US and I am going to get an intel quad and vista 64. Life is good. :)

02-18-2008, 12:57 PM
^yay! very cool! let us know how it goes!

02-18-2008, 10:50 PM
OK, put the finishing touches. Wound up with a Q6600, an Intel board, the 8 gigs of RAM, Vista 64 OEM, 2 Segate 1 T SATA HDs, and a BFG 8400 with 512. I'll slap it all into my current case.

When things get up and running I'll run a test on the object as well as a scene and report.

Yoo hoo... new toy :)

02-24-2008, 01:32 PM
Nice! Yeah, I'm dying to know, since those are the proc and RAM I want to install...

02-25-2008, 02:04 AM
Well, So far I have not been able to load Vista 64 to test the higher ram issue. There seems to be some kind of hardware conflict with the install program and I have not got any answers yet from the MS forum.

I may have to get a different HD and or DVD ROM.

I'll let you know as soon as I sort it out.