[CyberDildonics]

> precomputing stuff

Correct. In this context it means turning the geometry into frame by frame mesh data so that your animation isn't dependent on anything like bones, cloth simulation, muscle/skin simulation, soft or rigid body dynamics, inverse kinematics in the rig, expressions on the bones of the rig etc.

Not only does it break dependencies, but it is also easier to get that geometry into another part of the pipeline like effects, lighting and even compositing.

[solidangle]

So does a renderer such as RenderMan use a baked format like Alembic or a format like USD?

[joeld42]

Neither, and both. Lets say you have a scene with a cup of coffee spilling onto a desk...

The overall description of the scene, where the desk is, the lights and materials etc, could be stored in a USD scene.

Some tool would read this and generate an intermediate file in RIB format. This would be what a RenderMan renderer actually reads.

For the fluid sim itself, the generated RIB file might contain a reference to a plugin which points to the baked Alembic data, which would be a directory with one alembic file per-frame of mesh data representing the fluid surface.

[CyberDildonics]

The basic answer is it uses neither file format but what it does use has similar attributes to both.

RenderMan uses RIB files for its scene description and RIB files contain geometry. This means that one RIB file would typically contain a frame of data and source the shaders and textures from separate file paths.