In the last steps of v2's release, it was found there were some bungs in the old float 32 encoder (sometimes referred to as packing and unpacking), for unsigned precision. It was switched out with one that matched the excellent writeup from https://www.vizitsolutions.com/portfolio/webgl/gpgpu/speedBumps.html. Code from: https://www.vizitsolutions.com/portfolio/webgl/gpgpu/js/ToUnsignedBytes.js

What a nice solution!

Knowing this was likely a decision made in haste, it was thought to keep the legacy encoder, and offer the new one by default. This lead to a kernel setting 'useLegacyEncoder: boolean` to be added, after first checking that the new encoder worked with most standard hardware.