After possibly setting the new compression settings, each tile of the volume will be tested to see if it can be further compressed. If it can, it will be compressed.

All tiles are fully expanded into raw float data. This takes a lot of memory but ensures there are no special case constant tiles.

All tiles that are not constant tiles are fully expanded into raw float data. This is much more useful than the Uncompress option since constant tiles tend to be easier to support. This is mostly only useful if one needs the .bgeo to be read by another program/version of Houdini that lacks some of the compression options.

When tiles in the volume are compressed to constant tiles this will be the tolerance used to determine if the tile is constant. Zero will only allow lossless compression.

When compressing tiles by quantizing their values this is the tolerance to use. This can often be larger than the constant tile as some of the quality can be recovered by the dithering. Zero disables quantization.

When voxels are quantized to a lower bit depth some of the lost information can be recovered by adding noise to the pre-quantized voxels. This can be thought of as trading bit depth in the voxel with spatial resolution.

Dithering avoids banding in the resulting volume and, as volumes are often integrated for display, can keep the original weight of the volume.

Setting dithering to none will just do a pure thresholding operation. Dithering of ordered will use a constant dither matrix over each tile.

While 32 bit floats are still used for computation, whenever they are stored in the volume they are reduced to 16 bit floats. This cuts memory consumption in half at the cost of slightly higher computation costs and reduced precision of the floats.