Hi there, I was profiling some Modifier workflows and I noticed something in VTune. If you look at the attached image you’ll notice in the lower panel, with the graph, the second highlighted row is the CPU work of the modifier, there’s a lot of idle time in between, which looks like the time’s being used up by the upper highlighted row. When I expand that, it looks like it’s all GPU work (there’s a second attached image with that top row expanded). I was wondering if you knew what exactly was going on there, because I wondered if:
- Some of that work could be done on the CPU
- Maybe there’s a way to do the work asynchronously, where the data is prepared on the CPU and ‘fed’ to the GPU when needed (sort of like a cache)
