Hokay! We can render now! I figured out what was causing the Vram to spak out every time we tried to render a scene with more than one character in place.
2 things:
Texture size
Texture location
As fantastic as the ARNE server is for centralizing everything, it can't quite handle being used as a texture cache as well as a computer can by itself. The solution is dragging the project across to the scratch and rendering locally. The trick to making this work is making sure we move any updated files back to ARNE when we're done, and double checking the project settings each time we start up - which we do anyway...
I've resized all the textures down by an average of 50%. In some cases, even more. Now that the computer doesn't have to Unwrap, filter, light, layer and render 4 4096px square textures per character at 2 characters per screen, it can actually render a frame without crashing!
Here's a bit of maths for scale - an average scene before would calculate about 201326592 bytes of information per frame. That's 0.2 gigabytes (200Mb) of data streamed over a network into random access memory for textures only. Not including lighting, shadows, geometry or visibility calculations. All that cache data is processed to make the single .tga file, which is barely 7Mb...
Now, the same looking thing is only 12582912 bytes. That's only 12Mb (0.012gb) of information to play with for a render. Much easier. The only difference is the texture dimensions.
To prove my point, this is what a production render looks like. Ignore the dodgy reflections...
O____o
ReplyDelete