Investigate performance

[xd009642]

How can we get as much out of this as possible? Look at the channels, data representations etc. We're moving a lot of data around and potentially converting/cloning too much. Ultimately, this usually is a drop in the water versus the model running but for an infinitely fast model maybe there's some overhead we can eliminate