Komodo 10.4 released

[leavenfish]

I purchased the year long subscription a couple of weeks ago. Any insights/suggests on this would be appreciated:

98% of my usage is within Chessbase 14. Currently about 3/4 of that is analyzing individual openings I may play OTB or online. I've been rated 2000+ USCF for over 20 yrs. Right now for example I am going back to the system approach from A Strategic Opening Repertoire (Donaldson/Hansen)...updating some lines and looking for better choices in the mid/late opening area. Basically, I go to about move 18, turn Komodo on at a multi-PV of 3, and tab backwards until I notice evals moving up or down appreciatively. Then I investigate deeper making use of the 'x' feature to have the engine spot threats contained in certain moves my opponents might not see.

System: CPU: i7-6700 3.4 ghz - 16gb

QUESTION: What would likely be optimal settings for Komodo 10.4? As you presumably do something similar (?) when updating your opening lines (or on your opening books, of which I have your The Chess Advantage in Black and White), I would like to know what a 'power user' if you will of Komodo does.

Currently:
White Contempt: On at 10
Dynamism: 120
Variety: 0 (I figure I do not need this in analysis when looking at a multi PV of 3)
Hash size: 32
Everything else seems to be set at the default parameters.

NOTE: After I do this, I will throw all the games into a nifty little program called Chess Position Trainer and have it step backwards thru the repertoire and a deep depth evaluating each node. Then use the program to prune the resulting file and then use its unique training features.

[lkaufman]

I purchased the year long subscription a couple of weeks ago. Any insights/suggests on this would be appreciated:

98% of my usage is within Chessbase 14. Currently about 3/4 of that is analyzing individual openings I may play OTB or online. I've been rated 2000+ USCF for over 20 yrs. Right now for example I am going back to the system approach from A Strategic Opening Repertoire (Donaldson/Hansen)...updating some lines and looking for better choices in the mid/late opening area. Basically, I go to about move 18, turn Komodo on at a multi-PV of 3, and tab backwards until I notice evals moving up or down appreciatively. Then I investigate deeper making use of the 'x' feature to have the engine spot threats contained in certain moves my opponents might not see.

System: CPU: i7-6700 3.4 ghz - 16gb

Contempt should be zero for analysis in general.
Threads for a 4 core I7 should be 4 if Hyperthreading is off, perhaps 6 if it is on.
Hash should be way more than 32, maybe 768 or 1536.



QUESTION: What would likely be optimal settings for Komodo 10.4? As you presumably do something similar (?) when updating your opening lines (or on your opening books, of which I have your The Chess Advantage in Black and White), I would like to know what a 'power user' if you will of Komodo does.

Currently:
White Contempt: On at 10
Dynamism: 120
Variety: 0 (I figure I do not need this in analysis when looking at a multi PV of 3)
Hash size: 32
Everything else seems to be set at the default parameters.

NOTE: After I do this, I will throw all the games into a nifty little program called Chess Position Trainer and have it step backwards thru the repertoire and a deep depth evaluating each node. Then use the program to prune the resulting file and then use its unique training features.

[Werewolf]

Larry,
Could you have a go answering this question below?

I have an IDeA project with 350,000 positions in it, which were analysed with Komodo 9.1 @ 10 mins per position with each engine running on a single core.

Using the same hardware, how long would I have to analyse to get the same quality of analysis using Komodo 10.4?

[leavenfish]

To butt in a sec. You analyzed 350,000 positions for....6.65 yrs straight thru? Or is my math way off...?

[majortom]

To butt in a sec. You analyzed 350,000 positions for....6.65 yrs straight thru? Or is my math way off...?

Yes, I guess you calculated correctly, but for 1 core.
If there was used a few cores (threads) at the same time, we should divide by the number of cores (threads).

[Werewolf]

To butt in a sec. You analyzed 350,000 positions for....6.65 yrs straight thru? Or is my math way off...?

With IDeA you can analyse with many cores running many instances of an engine, but each engine runs on a single core. In total it took about 5 weeks.

[Cumnor]

Hi Carl,

If your going to change to K10.4 is there a simple way of changing a remote engine or do you spend days like me changing all the engines by hand.

At the moment I have 120 single core remote engines on my IDeA projects and I am rather reluctant to spend more days changing each one by hand.

[lkaufman]

Larry,
Could you have a go answering this question below?

I have an IDeA project with 350,000 positions in it, which were analysed with Komodo 9.1 @ 10 mins per position with each engine running on a single core.

Using the same hardware, how long would I have to analyse to get the same quality of analysis using Komodo 10.4?

Maybe something like 4 min per position.

[Lyudmil Tsvetkov]

Stockfish seems to prune/reduce more than Komodo even if we duplicate SF pruning and reduction rules, and I don't know why.

My latest theory is the equilibrium between evaluation parameters, that tend to find some moves and stick with them without much changes. So move ordering related to evaluation.

it should be sufficient that SF eval would favour more king attacking lines (by having a more extensive attacking eval with higher values).

when you go into a king-attacking sequence, chesswise more positions will lead with depth to a decisive outcome, or at least a pronounced evaluation edge, and thus exceed alpha-beta bounds and be prunable.

in quiet positions, it is more difficult to exceed those bounds, especially when an aspiration value is added. I guess duplicating SF pruning rules in Komodo would lead to less divergent behaviour in the endgame pruningwise?

similarly with pruning late moves: when you prefer a king-attacking sequence, only a couple of moves could strenthen the attack and be relevant, so late move pruning here should be safer to an environment of an engine that would pick more positional moves. No matter how good your move ordering, in quiet positions there is always a bigger number of reasonable moves, so late move pruning here is less safe.

PV-search, null move, etc. should be also more conducive in king attacking sequences for a plausible explanation of depth-reduction effect.

[Nordlandia]

Do principle of "Major Piece Redudancy" affects Komodo evaluation if there is two queens for one side, and many pieces remaining elsewhere on the board?

Penalty for the Queen pair (Larry Kaufman called it "principle of redundancy" although how much of penalty implied is unknown. In theory rook pair, knights and queens suffers from this.

Viktor Korthnjoi --->

Hypothesis "Having two Qs may be clearly worse than having a Q and 2 Rs. That would definitely be a special case.

During the analysis, I discovered something very remarkable: the board is simply too small for two Queens of the same color. They only get in each other's way. I realize that this might sound stupid, but I fully mean it. The advantage is much less than one would expect by counting material.

[d]1rbqnk2/4ppbp/p2p2p1/2p5/2P1PP2/3P2PP/1P1QN1B1/q1BNRRK1 w - - 0 1