I was able to render an incredible amount of tris using mayabatch, cache flushing and a proper relation between memory limit and task size:
Code:
MEM 0.0 info : heap size limit set to 640 MB
[...]
SCEN 0.0 progr: begin scene preprocessing for frame 1
SCEN 0.0 info : 148226 geometry leaf instances (148226 scheduled, 0 cached, 0 shared)
SCEN 0.0 info : 1 light leaf instance
SCEN 0.0 info : wallclock 0:00:08.69 for scene preprocessing
SCEN 0.0 info : allocated 294 MB, max resident 294 MB
RCB 0.0 info : scene extent: (-0.51,-0.51,-0.51) : (384.51,0.51,384.51)
RCB 0.0 info : scene diameter: 544.494920
RC 0.0 warn 082008: jittering disabled because max samples < 1
RC 0.0 info : option: scanline rapid
RC 0.0 info : option: samples collect 1
RC 0.0 info : option: shading samples 1.0
RC 0.0 info : option: samples motion 1
RC 0.0 info : option: trace off
RC 0.0 info : option: shadow off
RC 0.0 info : option: lightmap on
RC 0.0 info : option: motion off
RC 0.0 info : option: luminance weights 0.212671 0.71516 0.072169
RC 0.0 info : option: caustic off
RC 0.0 info : option: globillum off
RC 0.0 info : option: finalgather off
RC 0.0 info : option: filter clip mitchell 4 4
RC 0.0 info : option: render space object
RC 0.0 info : option: face both
RC 0.0 info : option: field off
RC 0.0 info : option: hair off
RC 0.0 info : option: hardware off
RC 0.0 info : option: task size 16
RC 0.0 info : option: pixel preview off
RC 0.0 info : option: lens on
RC 0.0 info : option: volume off
RC 0.0 info : option: geometry on
RC 0.0 info : option: displace off
RC 0.0 info : option: premultiply on
RC 0.0 info : option: colorclip raw
RC 0.0 info : option: output off
RC 0.0 info : option: merge off
RC 0.0 info : option: virtual framebuf on
RC 0.0 info : option: image type interpolate
RC 0.0 info : 0 rgba yes
RC 0.0 info : camera: focal length 1.37795
RC 0.0 info : camera: aperture 1.41732
RC 0.0 info : camera: aspect 1
RC 0.0 info : camera: resolution 512 512
RC 0.0 info : camera: clip 1 100000
RC 0.0 info : camera: frame 1 1 0
RC 0.0 progr: rendering
RCI 0.0 progr: begin intersection preprocessing
RCI 0.0 info : using the rasteriser for eye rays
RCI 0.0 info : not tracing secondary rays
RCI 0.0 progr: end intersection preprocessing
RCI 0.0 info : wallclock 0:00:00.00 for intersection prep.
RCI 0.0 info : allocated 351 MB, max resident 351 MB
PHEN 0.0 info : ----------------------------------------------
PHEN 0.0 info : mayabase version 7.5, compiled on Jul 19 2005.
PHEN 0.0 info : ----------------------------------------------
JOB 0.0 progr: 0.0% rendered on Freelance02.0
JOB 0.4 progr: 0.1% rendered on Freelance02.4
[...]
RC 0.0 info : wallclock 1:41:01.67 for rendering
RC 0.0 info : allocated 719 MB, max resident 733 MB
GAPM 0.0 info : triangle count excluding retessellation : 1195294464
GAPM 0.0 info : triangle count including retessellation : 1603373184
DB 0.0 info : disk swapping: 0 objects (0 KB) swapped out in 0.00 s,
DB 0.0 info : 0 objects (0 KB) swapped in in 0.00 s, max swap space used 0 KB, 101406812 KB f
ailed
MEM 0.1 progr: scene cache flushed asynchronously 247 MB for module JOB, now: 520 MB
MEM 0.0 info : 13647 scene cache flushes freed 99205 MB in 0.14s
MSG 0.0 info : wallclock 1:43:20.97 total
MSG 0.0 info : allocated 142 MB, max resident 733 MB
Press any key to continue . . .
From the maya UI, I noticed mr comes back with 'mental ray out of memory, aborting' when hitting the memory limit, instead of flushing the cache and going on.
I have a hard time finding the right correlation between memory amount and task size, there does not seem to exist some general rule (there is a very general rule, but it's too general to be applicable). Mr seems to have problems when many small objects fall into the same render tile. On the other hand the render times seem to raise exponentially with smaller (4, 8..) job sizes. I havent seen much info on all this stuff yet, not getting much feedback either, which lets me assume this is not a quite common issue. I'm still grateful for sharing any experience on this.