Sometimes I need to write out a sequence of geo, but each step is bottle necked. My CPUs never get over 6%.

How would I pipe this into a TOP network or TOP Geometry SOP to write out the geo?

I get that PDGs are super powerful, but for me they'd be most useful as a parallelizer.

If your geometry is defined in SOPs, you can use the ROP Geometry TOP node to write it out. It works a lot like the regular geometry ROP. On the TOP can choose to create work items for a single frame or for a frame range. TOPs will create one work item per frame.

By default the work items will cook in parallel based on the scheduling settings. If you geometry network is mostly single threaded, you can probably set the Local Scheduler's “Total Slots” parameter to “CPU count less one”. That way PDG will schedule a work item for each of your system's cores.

If your geometry only takes a few seconds to cook per frame, you may also want to consider using some of the Batching options on the ROP Geometry, found on the second tab of the ROP Geo TOP. You can for example group 10 frames of work into a single “batch” work item in TOPs. The UI will still display each of the subitem individually and downstream nodes can index them per frame, but those item will cook in one process. This can help reduce the overhead from spawning jobs and loading external files, such as the .hip itself or other supporting files. If you have enough frames (say 240 frames, which would be 24 batches), you can still achieve a high level of parallelism.

Awesome, that did it!
So straightforward. Thank you so much