I think I've hit some kind of performance bottleneck with the Tile Pattern COP. It turns out that setting Seamless divisions to 128 or more, puts a 100% load on a GPU (RTX 3070) for about half a minute. A mere change of a display flag on nodes down the stream, in a small sized network in which the Tile Pattern resides upstream, also initiates these calculations, even if Tile Pattern has already been cooked.

And if, God forbid, at that time the histogram is displayed in the Composite View pane, the full GPU load can last for several minutes, effectively blocking Houdini window.

Speaking of histogram, it seems to be recooking all upstream nodes each time I access it. Why can't it calculate the histogram just for the node with a display flag, which is already cooked?