I was wondering what is the amount of credits awarded per completed task based on?

It should be probably related to how complex was each task, which in turn is proportional to how long the computations took (assuming the same number of CPU cores were in use in each task), but that doesn't seem to be a case here.

Check out those two tasks while they are still present in statistics:
- task 8544325, took 128486 seconds (almost 35,5 hours), awarded 3233 credits,
- task 8543654, took 5570 seconds (only about 1,5 hours), awarded 5379 credits.
Plus both were using 8 cores of the CPU, all of them loaded near 100 % (yet there tend to run up to 8 programs that complete work at different speed, so for some time there are less and less cores used, until the last program quits and all of them are restarted again with different parameters), so the CPU time used should be about 8 times larger than the runtime. Even if the CPU time reported in both cases is equal to the runtime, this doesn't explain why a task which needed about 25 times more calculations to complete awarded less credits than the other task.

I was wondering what is the amount of credits awarded per completed task based on?

It should be probably related to how complex was each task, which in turn is proportional to how long the computations took (assuming the same number of CPU cores were in use in each task), but that doesn't seem to be a case here.

Check out those two tasks while they are still present in statistics:
- task 8544325, took 128486 seconds (almost 35,5 hours), awarded 3233 credits,
- task 8543654, took 5570 seconds (only about 1,5 hours), awarded 5379 credits.
Plus both were using 8 cores of the CPU, all of them loaded near 100 % (yet there tend to run up to 8 programs that complete work at different speed, so for some time there are less and less cores used, until the last program quits and all of them are restarted again with different parameters), so the CPU time used should be about 8 times larger than the runtime. Even if the CPU time reported in both cases is equal to the runtime, this doesn't explain why a task which needed about 25 times more calculations to complete awarded less credits than the other task.

This project uses the in-built BOINC crediting system called CreditNew.
Unfortunately it does not work that well. It is supposed to learn as you return work what to award based on some inbuilt system.
I find that it really can't handle and work out Multi-threaded work units (on any project), it seems to get confused and award credits for single thread work instead.
I have had work units run for a combined 500,000+ seconds and get a few hundred points, which is disappointing.
And then I have similar size work units give a few thousand points so it is not consistent which is the main fault of the system.

But it is what we have unless it is bi-passed as a number of other projects have done.
That is my take on it.
Conan