TalkChess.com

Komodo 10.3 has been released at komodochess.com. It appears to be our largest elo gain since the subscription model was begun. We estimate the elo gain at 2' + 1" to be about 30 points after about 10,000 games against Komodo 10.2 on one thread (Windows version), and due to mp improvement the gain seems to be even larger on 4 cores at this level (34 elo after 2650 games). Elo gains tend to decline with longer time limits and against unrelated opponents so keep that in mind; I'll predict +25 average on the rating lists. Most of the gains this time are due to improved search, though king and pawn endgame eval has been much improved. As usual, purchasers of Komodo 9 or later version get 20% discount, and for current subscribers it is free. We would probably have called it Komodo 11 if not for our agreement with ChessBase. We are hoping to regain the top spot on the CCRL 40/40 list, but time will tell.

I've been purchasing single releases, but the way you guys are pumping out improvements...I'm probably going to become a subscription customer real soon.

Thx and Congrats!

lkaufman wrote:Komodo 10.3 has been released at komodochess.com. It appears to be our largest elo gain since the subscription model was begun. We estimate the elo gain at 2' + 1" to be about 30 points after about 10,000 games against Komodo 10.2 on one thread (Windows version), and due to mp improvement the gain seems to be even larger on 4 cores at this level (34 elo after 2650 games). Elo gains tend to decline with longer time limits and against unrelated opponents so keep that in mind; I'll predict +25 average on the rating lists. Most of the gains this time are due to improved search, though king and pawn endgame eval has been much improved. As usual, purchasers of Komodo 9 or later version get 20% discount, and for current subscribers it is free. We would probably have called it Komodo 11 if not for our agreement with ChessBase. We are hoping to regain the top spot on the CCRL 40/40 list, but time will tell.

Wow! I had no idea the gain was going to be so much! I can understand you might call it K 11.0 if it werent for your CB contract. Hopefully this great improvement will help you on lists and also in the next TCEC. Normally i like to see eval improvements more to make komodo smarter and also stronger but i understand that search improvements are a huge part of K's strength as well. We dont want him searching any dumb moves and rather spend all its time on good moves.

Great work.

Sincerely,

Tim.

kranium wrote:I've been purchasing single releases, but the way you guys are pumping out improvements...I'm probably going to become a subscription customer real soon.

Thx and Congrats!

I think many like me have not the opportunity to spend this money !!! Why it is not expected to lower the cost and maybe this will INCREASE sales ... Of course I imagine that there are costs every time to test every single step of improvement, but I think with a lower price you can even think to counter the rampant piracy and make happy those who do not have much money to spend!
It 's just my opinion
Thank you!

Scacchista1977 wrote:

kranium wrote:I've been purchasing single releases, but the way you guys are pumping out improvements...I'm probably going to become a subscription customer real soon.

Thx and Congrats!

I think many like me have not the opportunity to spend this money !!! Why it is not expected to lower the cost and maybe this will INCREASE sales ... Of course I imagine that there are costs every time to test every single step of improvement, but I think with a lower price you can even think to counter the rampant piracy and make happy those who do not have much money to spend!
It 's just my opinion
Thank you!

That's why we have the subscription. If we do five releases per year (which has been the case so far roughly) and you keep renewing subscription, the cost is $16 per version. If each new version is at least ten elo stronger than the previous one (almost always true) and some offer new features, I think that's quite reasonable.

Great job. I really like the competition between chess engine programmers.

lkaufman wrote:Komodo 10.3 has been released at komodochess.com. It appears to be our largest elo gain since the subscription model was begun. We estimate the elo gain at 2' + 1" to be about 30 points after about 10,000 games against Komodo 10.2 on one thread (Windows version), and due to mp improvement the gain seems to be even larger on 4 cores at this level (34 elo after 2650 games)....

Nice gain.

Larry,

Nice job on the (time to depth) speed up. Impressive

" Most of the gains this time are due to improved search"

I have to say, this always puzzles me. Every iteration (of every engine) 'improves' search or something else. How...can one continue to 'improve' things like search? It's probably too complicated for a simple answer...but could you 'dumb it down' for me?

My most recent Komodo purchase was 8. Thinking about getting 10.3.

leavenfish wrote:" Most of the gains this time are due to improved search"

I have to say, this always puzzles me. Every iteration (of every engine) 'improves' search or something else. How...can one continue to 'improve' things like search? It's probably too complicated for a simple answer...but could you 'dumb it down' for me?

My most recent Komodo purchase was 8. Thinking about getting 10.3.

It's a question of which moves to examine more deeply, and which ones to give up on early, and also what order to look at the moves in. Just to give one example, all strong engines prune moves that appear to be blunders if fairly near the end of the search, but the definition of what is a "blunder" keeps changing from one version to the next. Another example: checks and captures generally deserve to be examined more deeply than quiet moves, but how much more deeply is often changed. The list is long. But the bottom line is that Komodo 10.3 on two threads is probably stronger than Komodo 8 on eight threads. In other words, to get the same improvement by upgrading your hardware you might have to spend 30 or 40 times as much money.