![]() Keep in mind that the render nodes MUST have access to each and every element contained withing the scene file, otherwise the frame/job will fail. ![]() ![]() Typically, you will want network storage for this task, as it avoids having to copy the scene files and associated elements around to each render node prior to executing the job. The primary I/O loads will occur at the beginning of the render and at the completion as scene files and elements are loaded into the renderer and then final renders are written out. I/O is not as big an issue, but it does need to be addressed. On a 12-core system, with 1GB scene files, running a single scene render, then you would want 24GB of RAM to be comfortable. You will want enough RAM per core to load the entire scene file and its elements, as well as leave some for the renderer and the operating system. So, ensure that each render slave has enough RAM available to each core/processor to avoid starvation. Scene files can be composed of a large number of elements. Mixing lower capability processors in with higher capability ones may cause the distributed render to actually slow down, unless you have a job manager that can account for this (more on that in a bit). One caveat here: the CPUs need to be as similar as possible when using distributed rendering. If you have 10 processors rendering the job, then your scene file will take (approximately) 10 minutes. If you have 100 frames per scene file, and each frame takes 5 minutes, then a single processor will render the entire scene file in (approximately) 500 minutes. ![]() I don't know the typical frame count of your scenes, but, the math is fairly linear. So, the more CPUs you can throw at the task, the faster your distributed jobs will complete. Rendering is primarily a function of CPU. Let address the issues/requirements in order here. So, I feel that I might have some expertise to lend here. The farm(s) that I (and Jeff A) build rendered pre-viz and final scenes for Sin City, The Adventures of Sharkboy and Lavagirl, Grindhouse, as well as several commercials and smaller projects. In my past professional capacity, I built render farms for 3D and 2D (comps). The main thing you really need to worry about in all this is that you have a full 1gb connection between the machines so that their uploads between them don't take all day. It is not cost effective or technically efficient but it will work best for your artists.and in the end that is what you're going for. If you cannot handle the mac issues as many can't then I would suggest getting each individual artist their own second station to do the rendering on. As long as you can put up with having to deal with Mac OS. I am not sure what it is but macs seem to be much better at processing visual data than their equally powered windows counterparts. I would highly recommend looking into getting a mac server for any sort of video or photo rendering. The upload takes roughly 6 hours now so we are thinking of getting a server to handle the job next.įor your situation your 3d rendering is much more processing intensive than our 2d ones, but the principles are mostly the same. Right now we have about 10 artists working daily and they run their scripts and upload daily as well. They all use Mac Pro's as their workstations, then we have another mac pro that handles all the scripts and resizing and whatnot. ![]() I've got a bunch of 2d photo editors who run scripts similar to what you are talking about. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |