Try MR with this advanced MIA Material. Haven't tried yet, but it says it speeds things up and the behavior is better.
http://felixgeremus.com/?p=108 2013/2/15 Arvid Björn <arvidbj...@gmail.com> > Mostly to avoid cores idling, with large tile sizes you could really be > wasting time, especially for something that only occupies a small part of > the image, or a really detailed patch that's left to last running on a > single core after everything else is finished. But it's part of my > optimization routine, so occasionally I increase the size. Never heard of > the extra sampling issue, can't imagine it would over-shadow the 1:8 ratio > of cores rendering simultaneously though, unless it's a scene with > extremely even detail level. > > > > On Fri, Feb 15, 2013 at 1:33 PM, Ed Manning <etmth...@gmail.com> wrote: > >> On Fri, Feb 15, 2013 at 6:45 AM, Arvid Björn <arvidbj...@gmail.com>wrote: >> >>> Sure, use unified sampling, min 1, max 100, quality 2 (in this case), >>> enable FG (default settings in this case), 16px tilesize, spiral pattern. >>> Use Architectural material on everything, set all samples everywhere to >>> 1. Enable AO in Architectural, set AO shadow color to black. >>> >>> >> Hi Arvid -- >> >> Why tile size 16? I ask because I often use this setting (all the time >> for previews, often for renders) in order to minimize the amount of idle >> proc time while the last couple buckets finish, but have been told lately >> that it's actually a time-waster since the renderer has to resample all the >> bucket edges. I'm not sure which helps/hurts more, although I guess maybe >> there'd be a way to figure out a rule of thumb based on bucket time vs. >> number of cores... >> > > --