pelos

more important than the percentage, is how many core are you really using

the sim you have might not be hard enough for the power of the cpu.

can also mean the distribution of points to solve is not needing additional compute power, example

if you have 32 core, and you only have 10 points, maybe those 10 points can be solve with just one cpu.

is there a particular reason why you want to use your own environment instead the one that comes with Houdini?

you can install python 3 in the pc,
use houdini python2 to run a python3 as a subprocess,
the python 3 can do the tensorflow and the python2 is just a runner

os.popen.subprocess(“python3 file.py -args….”)

or a vm with linux and put houdini, python, tensorflow etc… all the stuff you need, not something that i think is a great idea since the VM use a lot of Ram, that you might end up needing for houdini process.

this are just quick ideas that came to my mind.