per GPU, the most important indicators are, in my opinion:
Ater congestion rate
Number of GCH busy (112 GCH per Ater)
You will also find all KPIs you can also monitor per cell (TBF estab, LLC bits, etc) but i assume you’re already monitoring them at cell level. The GPU level is just an extra, but not “KPI” per say.
The CPU power budget is the most obscure indicator I’ve seen. I’ve no idea what it means, but I think I found something about the CPU power budget in a document describing the MFS. However, I’m not sure such an indicator is useful for us, radio engineers.