This is strictly notes related to Speed and multi-GPU:
Standalone 3.0 alpha 1 with 12 GPU
(1) Display choppiness - I am finding with 12 GPU that the view-port display has choppiness when attempting to pan zoom or rotate, and this choppiness is there even with subsampling 4x4. In experimenting, I find that as I reduce the # of GPUs towards 1, the display linearly smooths up such that when you then pan, zoom, rotate, etc., with 1 GPU it is very smooth But, as I increase active GPUs and get closer to 12, it gets linearly increasingly choppy as I pan, zoom, and rotate. I note that the choppiness/stutter/low frame rate-ish effect occurs despite the fact that I have a dedicated card for display not used in the rendering (Titan X).
Most relevant, the linear increase in display choppiness does not occur in 2X build, in any range between 1 and 12 active GPUs, it is always smooth.
(2) Rendering slower (unless you max parallel and max tile samples), expected per developer notes

- 6 sec on 2.32.2

- 18 sec on 3.0 Alpha 1 - default

- 8 sec on 3.0 Alpha 1 - Max parallel samples and tile samples
Win 10 Pro 64, Xeon E5-2687W v2 (8x 3.40GHz), G.Skill 64 GB DDR3-2400, ASRock X79 Extreme 11
Mobo: 1 Titan RTX, 1 Titan Xp
External: 6 Titan X Pascal, 2 GTX Titan X
Plugs: Enterprise