Distributed rendering options

Материал из Blender3D.

Перейти к: навигация, поиск

Auto Threads

This sets the number of rendering threads to be the number of processors reported by the operating system. For example 2 threads on a dual-core CPU, 8 threads on 2 quad-core CPU's. Auto-Threads is enabled by default and in most cases will give the best performance, allowing teams to work with the same blend files on different workstations.

The command line argument to use auto threads is

   blender -t 0

Image Naming

Blender now supports the use of the "#" character in the render output path to determine the position and length of the framenumber in an image filename.

Previously blender would use 5 digits for the frame number which meant most animations over 7 minutes would not have their frames sorted correctly in many encoders, filemanagers and graphics applications (including blender).

This is solved by allowing the user to define frame numbering position and length. eg.

   * /tmp/my_ani_##_small.png -> /tmp/my_ani_01_small.png, /tmp/my_ani_02_small.png.. etc
   * //anim######.exr -> //anim000001.exr, //anim000002.exr etc


   * Only # characters used in the filename will be replaced with the frame number.
   * The only the last group of "#" characters are used.
   * The maximum length of the frame number its 16 characters.

The default output names for images has not changed so only use this if you need to set the number size for your project. Touch & No Overwrite (simple distributed rendering)

The touch option will create the image it is rendering to as an empty file, overwriting it with the rendered frame on completion or removing if the render is canceled.

This provides a simple way to know if a file is being rendered or not.

When the "No Overwrite" option will skip rendering frames where the image name already exists, this can be used to re-render parts of an animation that have missing frames.

When the Touch and No Overwrite are enabled, multiple instances of blender can render the same .blend file to a network filesystem distributing task across many computers.

Each computer will skip the frames that others have touched until a free frame is found to touch and start rendering.

This is not a perfect renderfarm solution since its possible 2 computers will render the same frame if the network filesystem doesn't update fast enough. however on a fast LAN running SAMBA or NFS the chances of this happening are very low.

Личные инструменты