When I do WFO with multiple parallel cores, I recognised a recurring behaviour.
All instances running quite fast, but the last instance (not the master one) runs endlessly.
Despite the fact, I am running a 13900 K with 24 cores and I am using via Process Lasso the performance cores only, it doesnt change. Surely, the more parameter to optimize the greater the issue. Also RAM is not an issue (64GB).
I am using the Zorro64 and it seems to be even slower than Zorro32?
I am using NumCores variable with fix values or n-1 , no difference ?

Does anyone recognize the same behaviour?

M.

No. But if the last thread does not end, but the others do, maybe it's a script issue in the last wfo cycle. You can open the thread window and check the current state.

Yes, I always do it like this, but this behaviour I have in many scripts. This is why I am asking.
The last thread needs ages to finalize the training. It ends finally but it lasts 5 x 10 times longer than the other parallel threads.

I was playing around with NumCore. Despite the fact that I have 32 logical cores, I ran Zorro64 now with NumCore=3 and it is running a little bit more stable, but not good. I also using Process Lasso and allocate Zorro to the Performance cores of the Intel i9.
Actually, I feel this is a Win 11 thing and parallelizations seems to be an issue, when it get complex. Linux might be better prepared for this.

Anyone recognised similiar behaviours?

So the last thread terminates, but needs longer than the other threads? And this happens with any number of cores? Is that correct?

Yes, what I see is that the last thread needs endlessly . Process Lasso or Task manager showing that this last instance consumes much less RAM than the others, but the CPU consumption is even slightly more than the others.

I am running a strategy using the python bridge and calculating HMM (Hidden Markov Process) and PDA (Principal Component Analysis). What I am wondering, is that the other instances running fast.

I tried different number of cores:

At the moment,I am using Zorro only on the Intel i9 performance cores, but nothing has changed. I am doing this strategy with NumWFOCycle=5 and for each parallel run there is an on isolated core, no other activities . I put the other activities (noin Zoroo/Python) to Core 17 to 32.