View Full Version : 3DSMax Render Speed - Scanline vs MR
03-04-2011, 11:05 AM
I've always used the "default Scanline" renderer in max for everything including baking AO and normalmaps. However, I've been doing a little research and I've seen conflicting opinions that MR is faster for basic rendering tasks like these (especially with multi-core CPUs). I just wanted to ask if there is much truth in this - I've recently found that baking normalmaps in max from ZBrush'd models (even optimized down to sub-1mil tri counts) takes an absurdly long time and I can't fathom why. I'm wondering if Scanline might be the reason.
Any insight or links would be great, thanks.
03-04-2011, 11:30 AM
It is indeed true, “multi bucket render” or "Multi core rendering" will increase the speed of your render.Of course speed will come down to your settings but it will be probably faster using Mental Ray.
03-04-2011, 12:18 PM
I think its Scanline showing its age. I find lightracer very slow on highpoly models. that said for sub million i havent had any problems.. but i tend to only bake normals and colour in scanline. AO i do in MR as you get much finer control
03-04-2011, 12:33 PM
I've tried disabling light tracer but it doesn't seem to make much difference. The real problem I'm encountering is that anything which comes into max from ZBrush just takes forever to bake - eg a ZB mesh optimized down to 450k will take about 10x longer to bake in max, than an equivalent 450k subdivided object in max.
03-04-2011, 04:07 PM
marks, if you have a file available, upload it and i can check it for you. i do 10million+poly bakes with no problem, and max will spit out a 2k normal map in about 2-5 minutes. and about 15 min or so for a quality mr ao bake,
to address some specifics,
if your using go-z keep in mind max applies a direct x material to your hi-poly object.
i haven't tested what the result of baking with this applied is, but swap to default materials to be sure.
i would avoid rendering normal maps in mentalray, the sampling quality (more specifically the aliasing where meshes intersect) needed to match the scan line, makes its normal map renders much longer. but it is faster at low sampling rates. but generally and for consistency i always use scan line with the hammerssly sampler for nm.
for AO, i use mental ray, i just like the control and results i get, if your AO bake is going slow, try using the ao shader in the diffuse slot then just render a diffuse bake.hopefully this helps a little, and post your file if you wan't someone to look at it.
03-04-2011, 04:20 PM
Thanks for the info, nice to know that I won't need to bother messing with MR for bakes. I've been using scanline and hammersley since I started. I've been importing/exporting OBJ between apps, not using go-z. Everything has default materials applied to it, I even merged the 2 objects into a new scene and the problem persists.
I'm not even sure this is a hardware issue because I've been baking anything from 2-4mil in the last year on the same machine with no issues (although these meshes were modelled totally in max using a cage mesh and subdivided). The only similar issue I've run into previously was baking from 5.5mil+ and this was literally 3dsmax crashing due to running out of RAM.
03-04-2011, 06:17 PM
03-04-2011, 06:55 PM
bad quality bakes, I've tried it already.
03-04-2011, 06:59 PM
they never came out bad for me, maybe diff versions or settings or something.
Though, i never really used it on the kind of mesh you're describing, i mean, one with slightly less than what youd expect of a completely sculpted object. I mostly did it on characters and very high density stuff. They always came out fine for me.
03-04-2011, 07:27 PM
I had similar experiences with my short test of Xnormal. Maxs supersampler kicked its ass. What settings do you use?
vBulletin® v3.8.4, Copyright ©2000-2013, Jelsoft Enterprises Ltd.