Hi,

When running the same HDA on the same data through Houdini Engine I'm getting up to 10x worse performance than running directly in Houdini. This happens for a massively multithreaded part of calculations inside a compile block. When running in Houdini I'm getting around 0.6s for a run, while with Houdini Engine I can get between 0.6s a 6.0s, and the time can vary greatly between consecutive runs within the same session.

I tried bumping the HARS process priority and it seems to help a little, but it still is not consistent nor on par with the performance in Houdini.

My setup is:
- Windows 11 Business (10.0.26100 Build 26100)
- Intel Core i9-14900KF, 3200MHz
- GeForce RTX 5080 16GB
- 128GB RAM

Does anybody have any advice on this matter?

It appears that the issue was somehow connected to an 'overuse' of the compile block. When I limited compile blocks to parts of the HDA that benefited the most from multithreading, the issue disappeared, or at least is not as apparent anymore.