Hey Chad, 1. With Redshift it will roll over (out-of-core) to system ram once the vram is filled, which does slow the performance down (though its still faster than cpu rendering). As an example when I was comparing a 780 (3Gb) to a titan (6Gb) on smaller scenes they were fairly evenly matched, once we added something memory consuming like hair the titan was often 2 to 3 times faster.
2. I'm sure RR7 added something to address per gpu render jobs, Deadline certainly can. Tim - A man's cup of tea is no joke ;) On 7 August 2015 at 03:18, Chad Briggs <chad_fo...@elementxcreative.com> wrote: > We are eyeballing a new farm next year (about that time again) and this > thread has been great. I do have a few questions as we are currently using > arnold for everything and don't have much experience with Redshift. > > 1. everyone keeps chiming in on how you want the graphics card to have > as much memory as possible. What happens when you run out of VRAM during a > render? Does it just crap out? Or get really slow? > 2. Some folks were mentioning they assign one render job per GPU > rather than per node? Does your farm software have to be set up to handle > that? (we use Royal Render). > > Chad Briggs > Element X > > >
