Okay so Ive been chatting to one of my artists who thinks it would be very beneficial for the swarm renderer to work with x86 architecture (and also GPUs) to dramatically speed up lightmass times by using coprocessors such as the Intel Xeon Phi next to a Nvidia Titans.
I can certainly see the benefits of this myself as most of the fasted benchmarked systems in time based renderers are held by titan/phi combos.
Hi Daniel. I’m not sure what would have to be done to support Xeon Phi systems but Lightmass is already x86 compatible and in fact that’s the only architecture it can run on. Making Lightmass work on GPU’s basically requires a complete rewrite. Task-parallel algorithms don’t execute efficiently on GPU’s and Lightmass makes heavy use of task parallelism.
Ahh yes I did see the thread about that and the concerns there. I shouldnt have made the assumption because Rocket is 64bit that Lightmass/Swarm is also. That does bring up the question about lightmass being limited to 32bit memory addresses though or is that a non-issue when running 32bit applications on a 64bit system?
I think the Xeon Phi runs along side the CPU so it its possible it shows up as more cores in windows task manager (if it even runs on windows). I just thought it was interesting because I have seen the xeon phi/titan combos topping prerender benchmarks. It could certainly open some doors for smaller studios wanting to do bigger scenes without the need for render farms, cool.
Well, to make Lightmass ArchViz projetcs is good buying a Xeon Phi Systems to speed up the calculations?