Out-ofCore experiments with OCDS and Standalone 2.21.1
Posted: Mon Feb 16, 2015 11:40 pm
My primary render GPU is a 780ti with 3Gb and while this works out most of the time nicely with some careful texture management I do run occasionally into a case where I do need more memory. Thus I was happy to see the implementation of out-of-core rendering (i.e. using the main ram for extra textures) in 2.21.1 standalone. One can of course load things directly into Octane standalone via .obj import, but I find it easier to use OcDS for the scene setup. Hence I experimented a bit with .ocs export from OcDS and rendering in Octane 2.21. For .ocs export the scene needs to be first loaded in GPU memory as far as I can tell (I does not work if one does not do that) and thus we cannot directly set up a large scene and export. Instead I set the scene up in several parts (three in the example below, the two figures and the scenery), set up materials, and saved each part in the correct position relative to each other as .ocs. Then I fired up Octane Standalone 2.21.1 and loaded the scenery and imported the other .ocs files into this scene. One needs then to connect the geometry from the additional .ocs files to the first Render Target (via the Geometry Root node) so that everything appears in one image. To make this easier and allow for small adjustments in the placement, I exported the figures as .obj from DS and re-imported them to DS prior to setting up the Octane materials in OcDS and saving as .ocs. That way they appear as a single mesh in OcDS, making it easier to move them via the Placement node and to connect them to the render target (only one connection for each figure rather than one for each submesh). Might sound like a lot of work, but is actually quite quick. Then I enabled out of core rendering and allocated 8 Gb of ram to Octane. And boom it went with out issues. This scene used 2.5 Gb on my GPU and a bit over 1 Gb of main ram. It does render slower which is not surprising due to the need to move textures in and out of the GPU, but for this particular example it was only about 5% slower than the same scene with one character which fit into my GPU. Below is the finished render. It is still a work in progress, so materials are not fully tuned. But it worked! Hopefully this feature will at some point be available directly via OcDS. It is very useful if you need to squeeze just a bit more into a scene...
Ciao
TD

Ciao
TD