Re: maximum polycount environment variable?

["peter_b" <peter_b(at)skynet.be>]

Hi Ian, and those who care:

 

I tried cutting the object in pieces, and indeed I can get higher detail then,

with again a maximum of 2 million tris per piece/object at rendertime.

 

I understand that there's a limit to avoid creating ridiculous amounts of data and crashing the software, but even with just 1gb of ram, more than 1 million polys can be handled, I did about 5M with this test now.

and nowadays systems with 2gb and up are common, so this limit seems outdated to me.

I know its solved in v5.0 - but if there's an env variable, to push this limit for v4.2, I would really love to know.

splitting objects manually is not always easy/possible. In my test I can clearly see the seems...

 

 

 

 

 

 

I am simultaneously asking mental if this env var still exists; but I was speaking to them about this exact issue two days ago, and they said that the heuristic was not visible to the user.  So is a hidden env variable visible to the user ?  ;-)

 

-----Original Message-----
From: owner-xsi(at)Softimage.COM [mailto:owner-xsi(at)Softimage.COM]On Behalf Of peter_b
Posted At: October 27, 2005 1:32 PM
Posted To: xsi
Conversation: maximum polycount environment variable?
Subject: Re: maximum polycount environment variable?

thanks Ian, Ill be looking into it in the morning.

it seems like manually splitting will be my best option.

 

for my polycount: the allready subdivided object was 40k polys.

I think with ~4 million tris, I ll be where I want.

 

thanks

 

 

As I mentioned in the other fork, mental has an internal heuristic that reduces the triangulation count.  Try cutting your model in two, and apply the same settings.  I think you will find that the two objects separately will have 2 million each (unless their heuristic caps the number of triangles/poly... but it is still worth testing).

For parametric, and I believe LDA as well, mental takes one object and produces a single displaced object internally.  This is why their heurstic is there, they don't want to create too large an object since it would require many megabytes of continuous data (a big problem on 32 bit machines).

Fine displacement, on the other hand, can produce what are referred to as "split objects" (what in 5.0 we create automatically as well).  Basically, your initial object may lead to many split objects, each with a "reasonable" number of triangles in them.  Hence, no heuristic is necessary (except good luck with your 40k * 16 for two subd levels * 4^15 = 687 trillion quads?... there's a reason we cap it at 7!)

Yes, fine displacement does presampling for perhaps ages, but the reason it does that is to better allocate the triangles, hopefully giving better bang for your buck.

And yes, 2 million with fine displacement is too low a limit, but you are not using that -- for LDA, it is more reasonable.

 

-----Original Message-----
From: owner-xsi(at)Softimage.COM [mailto:owner-xsi(at)Softimage.COM]On Behalf Of peter_b
Posted At: October 27, 2005 12:47 PM
Posted To: xsi
Conversation: maximum polycount environment variable?
Subject: Re: maximum polycount environment variable?

 

i cant imagine capping out at 2 million with fine displacement.

 

you mean, that it sounds too low as a limit? thats why I wrote here, Im sure I've had higher counts before. Not on one object though.

are you sure you are not setting the max displacement? that will limit the amount of geometry made by clipping displacement at that value?

under most conditions i use fine, view dependant.

like I explained in another mail, fine first starts to do displacement presampling for ages.

with LDA is doesnt.

it displaces the mesh in a minute and a half, then "computing camera volumes" for 2 minutes (I wonder if this is the actual creation of the BSP tree?) and then renders in 10-15 minutes. in the other mail I mentioned the parameter settings, but I've tried with and without view dependent, raising the subdivisions up to 15 (with LDA it does that!) lowering the treshold to .1 with view dependent.

That should make it create a huge amount of polys. But it doesnt.

 

 

 

_sam

SAM CUTTRISS
3D AFICIONADO

peter_b wrote:

cant find it in the archives, and I know it existed in soft3d

 

so is there an environment variable for XSI or mental ray that limits the size of objects?

or that splits up objects at rendertime or ....

 

Im trying to push displacement subdivisions on this object, and it doesnt want to go over 2 million tris.

reducing the length treshold weirdly gives me less tris (something around 700 000) and is much less detailed.

 

any pointers?