Buffers and Uniforms

The differences between Buffers and Uniforms and their distinct usages.

1. Usage

In the graph system, some pins are square and some pins are round. This shows whether the data is what we call a Buffer, or a Uniform.

• A buffer, which has a square pin, carries data that can be different from one position to the next.

• A uniform, which has a traditional round pin, is used when a value is the same for every position.

Some pins are uniform (round), others are buffers (square), and some pins (grey center) can be either.

Uniforms can be plugged into any input pin, including Buffer pins. Buffers can specifically only be plugged into other Buffer pins. They cannot be plugged into a Uniform pin. Some pins can be converted from one to the other by right-clicking the pin and clicking the "convert" option in the context menu.

2. Technical Details

A single-length Buffer is functionally identical to a Uniform, and using one or the other will not have any impact on performance. Performance considerations only change if the Buffer contains multiple values, for instance because it uses Position somewhere in the node-chain to calculate a value.

The most straight-forward rule of thumb for whether a given value should be a Bufferor a uniform is this:

• If a value changes based on position, it is almost always a Buffer.

• If a value is the same for every position, it is almost always a Uniform.

On a technical level, a Buffer is simply a stack of data: a single Buffer pin represents the data for every point within a chunk at the same time, rather than just being a single value. Having this data in a stack allows it to be operated on in bulk, i.e. by calculating many values at once using ISPC on the CPU, or (not currently supported, but theoretically possible) GPU compute.

Uniform values are constant throughout all of a chunk's positions. If needed, a Uniform value can change from once chunk to the next. For example, the Get LOD node returns a uniform. It reports the current LOD, which varies from one chunk to the next. Because the LOD value will never change within a chunk, however, it can still be a uniform.

Additionally, there are a handful of special types, such as Surfaces and Point Sets, which are treated as uniform under the hood, despite them being intuitively 'different per position'. In these cases, the uniform is actually a struct that contains a set of buffers. The entire struct is uniform per chunk, but this single uniform struct holds a value for each position within its chunk inside the buffer elements. To avoid confusion, these types have visually distinct icons and are not presented to the user as uniforms.