Hello,

I recently built an HDA for our project to simplify the process of caching by adding a couple of features to a filecache. The main feature I intended to implement was a background render through TOPs (With a couple of extra features). Unfortunately, it is a lot slower compared to the normal FG-render, so I assume it is not utilizing the PC resources properly. My tests that excluded all extra features of the TOPs-Render had the exact same result, a much slower simulation (~4-fold time taken). I configured the local scheduler to use all cores except one and set it to process only one item at a time.

Here is the PDG-tree:

(Side note for the partitionbyframe-TOP: that is there to debug a crashing issue with updating the comment.)

And here are the two output directories. On the date modified you can see the difference in speed.



Can you point me in a direction of what I might do against this or is this a conceptual misuse of mine for PDG?

Thanks,
Martin

If the output_filecache node is a simulation, make sure you have “All Frames in One Batch” enabled. Also, note that PDG runs work its out of process. This means that if the work items are very fast to cook then the cost of spawning a process takes up a lot of extra time relative to the actual cook. In those cases you may also want to enable batching since it'll multiple frames in the same process.

Thanks for the advice. I forgot to mention I was fully aware of that already and have that enabled so it runs in one process. Yet, the simulation is still much slower. Anything else I could try?

What version of houdini is this? Can you tell by looking at your O/S performance monitor if there is a difference in the core utilization? You could try enabling the Houdini Max Threads parm and setting it to 0 or -1.