just wondering if its possible to speed up the rendering process when using matte objects
it becomes exponentially slower
any geo with over 1million poly (have 10million poly hair hold out right now)
it just becomes painfully slow, how come?
also when using hundreds of individual objects krakatoa doesnt seem to like either
anything we can do to speed this up?
just finished render
48minutes and 11 seconds 
38k particles with a 10million poly matte holdout
You could render a Z-depth pass in another renderer (like Scanline, VRay, mr) as a sequence and load the sequence as “Initial Depth Map” in the Matte Objects rollout.
Krakatoa performs the same kind of rendering using its own Scanline renderer, but it sounds like it is too slow with large amounts of geometry.
Also, are you using Motion Blur?
i did try to use a zdepth pass the lighters made in maya once but that did not work too well
maybe ill try again to make my own
jes, with motion blur 
ok the camaera motion blur was enbaled, so with that turned off, and without kraka motion blur, render took 4minutes as opposed to 4 seconds 
does it not “just” have to load the geometry into memory, and then occlude the particles?
would be interesting comparison, to try and dim the density down to 0 of all particles occluding the object pointing to camera
so if particle a is behind geometry (facing camera) then assign 0 density
wonder if that would be faster in a magmaflow on the prt loader, than to use matte object
That approach tends to leave gaps around the edges of matte objects where the center of the particle was occluded but some of the pixels were not actually behind the geometry.
The Matte Objects are rendered into a Depth Map.
So no, they are not “just loaded into memory”, they are actually rendered in a similar way to what Scanline renderer does, including evaluation of their material and taking into account its opacity mapping, so 100% Transparency from a texture produces a hole in the mesh that shows particles through it. Thus, the shader assigned to the Matte Object could also play a role in the slowdown.
A simple test I made:
1 Million Particles : 1.9 seconds to render.
1 Geosphere with no material, 800K faces (200 segs) as Matte: 11.418 seconds to render.
16 Geospheres, 50K faces, 800K faces total: 11.73 seconds to render.
16 Geospheres, 200K faces, 3200K faces total: 34.856 seconds.
16 Geospheres, 800K faces, 12.8M faces total: 132.53 seconds.
Seems to be scaling nearly linearly, as I would expect it to.
Using Magma would perform raytracing. With significant amounts of particles (1M+), this approach will be slower.
In fact, Krakatoa Mattes still support raytraced mattes internally when rendering with a Spherical lens camera override, because that cannot produce good depth map.
I switched the camera to Spherical and rendered 1MP against the 16 Geospheres with 50K faces each - got 23.398 seconds instead of 11.73s. The other two tests produced 67.158s and 290.12s.
As you can see, the raytracing was twice as slow with 1MP, and would get even slower with increasing particle count.
Also, raytracing does not allow for anti-aliasing, while the Supersampling of the Depth Map can find partially occluded particles.
If Motion Blur is enabled, the Depth Map will have to be regenerated on each pass, which obviously multiplies the times. If geometry is deforming, you have some ways to control how often it will be evaluated and given consistent topology, you could interpolate a bit, trading off time for lower quality results.
There are some things we can optimize in the future, but I think the times you saw were caused by a combination of settings that forced the reevaluation of the geometry many times.