Komodo/Chessbase allows to set the number of lines that are calculated in parallel.

(See here:

http://help.chessbase.com/Komodo/15/...tml?000001.htm

under General operation -> Program windows -> Engine window)

Is there an optimal setting? I assume it might depend on the number of cores assigned to the engine. Does anyone know more about this? Google couldn't help me.

The general rule with any chess engine is to set it to calculate only as many lines (so-called principal variations, PVs) as there are plausible moves in the particular position. Remember that the time needed to calculate to a certain depth is almost proportional to the number of PVs.

If you see early during the calculation that there are fewer good candidate moves than there are PVs, then you may want to reduce the number of lines to get higher speed.

If your RAM is small, then the size of the hashtable (HT) also matters - if you tell the engine to calculate too many PVs, then the HT will be filled up too fast, which means that the engine will start deleting already evaluated potential positions from its memory to give room for newly evaluated ones. That said, as far as I understand, the probability of 'forgetting' an important line during such replacement is quite low, and the effect is usually insignificant.

Thank you for your post.

I use 8 (out of 16 GB RAM) for the hashtable, 4 cores and usually between 8-10 lines in parallel.

There are often a lot of "plausible moves". I have seen plenty of positions where the computer evaluates ten different moves within 0.20 or less. Should I increase the number of lines for those positions? How big has the difference in evaluation move have be between the best and the nth best* to limit the lines to n?

*or the nth and (n-1)th best

I have no idea, it depends on the position

If you're studying openings, just use a database of ICCF games - correspondence games are a product of a lot of CPU hours of Komodo / Stockfish / Houdini evaluation, so you don't need to rerun it.

Or maybe get a modern Nvidia GPU and use Leela Chess Zero for opening evaluation - Leela has been trained on an enormous number of games, and a specific opening position comes up more often than a specific middlegame one, so neural networks are very good at openings and early middlegames.

Mark Lefler, the main Komodo developer, once confessed on the chess.com 'computerchess' Twitch chat, or maybe on the TCEC Twitch too, that he admired Leela's ability to evaluate openings. Also, he's announced that he'd been looking into the possibility of implementing a neural network in Komodo.