How many plys without pruning?

[Ralf Müller]

Hi,
I'm developing a new engine and I'm curious about how many plys an engine can calculate just with iterative alpha-beta-search in ca. 3 seconds from the start position.

Thanks!

[AlvaroBegue]

I just checked with my new [still nameless] engine and I got depth 11.

[Ralf Müller]

Wow, so much!
On which hardware? With or without QS Search?

So depth 11 you reach only per efficient move ordering? I'm at depth 6... *being embarassed*

[AlvaroBegue]

It's my 1-year-old laptop, which has an i5 processor. This is with quiescence search, although I only look at captures that have SEE>0.

I use hash tables and I sort moves like this:
* Hash move
* Captures in MVV/LVA order
* Killer move
* Non-captures in history-heuristic order (I increment entries by 2^depth)

I also use PVS, but even disabling it I also get to depth 11.

Part of the trick is my engine is searching over 3 million nodes per second.

[Ralf Müller]

Many thanks! Is it possible for you to test how many plys you get without move ordering?
I wonder where I lose so many plys...

[AlvaroBegue]

Well, "without move ordering" depends a lot on the details of what order comes out of your move generation.

Try to add MVV/LVA order to the captures, because that's huge, especially in QS.

[Ralf Müller]

My depth 6 is without QS ^^ I know, I'm a bad programmer... Do you still have hope, that it lacks on the move ordering or is there propably a bad representation/generation or something other?

[hgm]

Tests without move ordering are pretty meaningless. It would be better to just compare nodes per second. If your nps is OK, but your depth sucks (without reductions), you know the problem must be move ordering. Between 500 knps and 1 Mnps should be workable. If it is as low as 50 knps you'd better work some more on it. In the end it will not matter very much if you do 500 knps or 2 Mnps. Perhaps 100 Elo. While with move ordering and reductions you can earn much more than 1000 Elo.

(But probably you should first make a QS, Without QS the search will be very erratic, and it will be difficult to get a good move ordering. Because what is good for one depth (a PV ending in QxB, might be disastrous at the next depth, when it turns out the B was protected...)

[AlvaroBegue]

If you can get a working chess program, you are not a bad programmer.

[Ralf Müller]

I have 150,000 - 400,000 evaluations per second.
So my move generation is the problem?