Here are all the new additions and updates!

Core

• Output VDBs have their VDB type intrinsics set.
• The solver has substep parity. The results of simulating for twice as long and retiming to double the speed is the same as simulating with 2 substeps and double the time scale. The exception is if you are using turbulence because it uses the actual frame number to compute the noise pattern. You can get a similar feel by adjusting the speed of the turbulence accordingly.
• New velocity stage output option. The solver can output the velocity that is fully corrected post pressure solver. This makes retiming and particle advection much more stable. Previously the solver could only output the advected velocity field that was not as suitable for these operations.
• Many small performance optimizations.
• Fixed a bug where the maximum Jacobi iterations parameter wasn’t clamping the correct value.
• Fixed a bug that may have caused issues related to advection. This does mean that an Axiom 3.1 sim will look slightly different compared to an Axiom 3.0 sim.

API & OS Support

• 5-25% faster performance when using Metal.
• 5-25% faster performance when using OpenCL on macOS.
• More stable Cuda support.
• New solvers will default to Metal when using macOS and Cuda if you have an Nvidia GPU on node creation.
• Improved error reporting when a device in unable to be used.

Forces

• New fast perlin noise type. It’s ~5x faster than standard perlin however has limited controls. Solver still uses standard perlin by default.
• Fixed bug with dissipation not respecting the time step menu.
• Fixed bug with disturbance that caused sub frames to use the same seed as the whole frame. Now each sub frame get's a unique noise pattern.
• Fixed a bug with simplex noise in the turbulence force that caused it to be non-zero centered.