Best practice for setting draws, tune, chains & cores

Are there any best practices when deciding on number of draws, tune, chains & samples?

I know that I should adjust tune so there’s time to converge before samling. And I know that I should have at least two chains. But beyond that?

Is it better to do more chains with fewer draws or vice versa?

Currently I’m running as many chains as cores, but only 1000 draws each.

1000 draws is usually sufficient with a well tuned HMC/NUTS for most application, otherwise, the rules of thumb is if you want 1 digit more precision of your estimate you need to 10x your sample size (1000 draw → .001 precision, and technically 1000 effective samples)

As for number of tuning samples, I am quite impatient so 1000 is usually the maximum I do