What are Stockfish and Komodo estimated rating for these 5 depths ?

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
lkaufman
Posts: 4286
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: What are Stockfish and Komodo estimated rating for these 5 depths ?

Post by lkaufman » Sat Jul 25, 2020 3:04 am

Ovyron wrote:
Sat Jul 25, 2020 2:50 am
Milos wrote:
Sat Jul 25, 2020 1:32 am
As a matter of fact I doubt you even know how to set up GUI to play SF at depth 1. You are probably playing nodes=1.
Are we even in the same world? Around here Stockfish 8 Depth 1 gets its queen trapped with ease:



If that's FIDE 1500 level then I don't know what FIDE ELO is. I can beat Stockfish depth 1 100% of the time...
I'm sure that 1500 is too high an estimate for 1 ply given that the total time limit for the game is say one hour (of which SF will use under 1 second), but it might be accurate for say a ten minute total time limit. Of course you can beat SF 100% of the time, but I'm pretty sure you are strong enough to beat 1500 rated humans 100% of the time too, so this doesn't resolve the question. In the 1980s machines that did 4 ply full width (plus qsearch) got ratings like 1600 or so in standard tournaments, and today's 1 ply searches, though "smarter", are nowhere near the level of a 4 ply search with even fairly basic chess knowledge of the mid 1980s. But in blitz, humans rated 1500 just blunder pieces left and right. We have to specify the total game time assumed for the rating.
Komodo rules!

Milos
Posts: 3762
Joined: Wed Nov 25, 2009 12:47 am

Re: What are Stockfish and Komodo estimated rating for these 5 depths ?

Post by Milos » Sat Jul 25, 2020 3:10 am

Ovyron wrote:
Sat Jul 25, 2020 2:50 am
Milos wrote:
Sat Jul 25, 2020 1:32 am
As a matter of fact I doubt you even know how to set up GUI to play SF at depth 1. You are probably playing nodes=1.
Are we even in the same world? Around here Stockfish 8 Depth 1 gets its queen trapped with ease:



If that's FIDE 1500 level then I don't know what FIDE ELO is. I can beat Stockfish depth 1 100% of the time...
As usual you are talking bollocks.
Current SFdev is like 150+Elo stronger than SF8 at depth 1. What's next you are gonna play against SF1.5.1 and claim that you can beat it blindfolded???
And even monkey can beat it 100% of time since fixed depth games are 100% deterministic, so all one has to do is memorize the moves.
I can bet you wouldn't even be able to beat SFdev depth 1 with 6 moves limited randomized book in 10 games match.

Milos
Posts: 3762
Joined: Wed Nov 25, 2009 12:47 am

Re: What are Stockfish and Komodo estimated rating for these 5 depths ?

Post by Milos » Sat Jul 25, 2020 3:15 am

lkaufman wrote:
Sat Jul 25, 2020 3:04 am
In the 1980s machines that did 4 ply full width (plus qsearch) got ratings like 1600 or so in standard tournaments, and today's 1 ply searches, though "smarter", are nowhere near the level of a 4 ply search with even fairly basic chess knowledge of the mid 1980s.
Again you are just inventing stuff. Not the first time. Care to document your wild claims?
Are you even aware how much more SF extends in qs than those early programs? Just looking at nodes search SFdev 1ply goes over similar number of nodes as full width 4 ply negamax. And not getting into how much more tuned SFdev eval is just for material+PSQT (that btw. didn't even exist in 80s), not to mention king safety, all bonuses, endgame knowledge, etc. etc.
SFdev 1 ply is hundreds of Elo stronger than '80s program 4 ply search.

Chessqueen
Posts: 1032
Joined: Wed Sep 05, 2018 12:16 am
Full name: Nancy M Pichardo

Re: What are Stockfish and Komodo estimated rating for these 5 depths ?

Post by Chessqueen » Sat Jul 25, 2020 3:26 am

Milos wrote:
Sat Jul 25, 2020 3:10 am
Ovyron wrote:
Sat Jul 25, 2020 2:50 am
Milos wrote:
Sat Jul 25, 2020 1:32 am
As a matter of fact I doubt you even know how to set up GUI to play SF at depth 1. You are probably playing nodes=1.
Are we even in the same world? Around here Stockfish 8 Depth 1 gets its queen trapped with ease:



If that's FIDE 1500 level then I don't know what FIDE ELO is. I can beat Stockfish depth 1 100% of the time...
As usual you are talking bollocks.
Current SFdev is like 150+Elo stronger than SF8 at depth 1. What's next you are gonna play against SF1.5.1 and claim that you can beat it blindfolded???
And even monkey can beat it 100% of time since fixed depth games are 100% deterministic, so all one has to do is memorize the moves.
I can bet you wouldn't even be able to beat SFdev depth 1 with 6 moves limited randomized book in 10 games match.
Stockfish 8 :shock: You should get at least Komodo 11 which is Free, with time control of 15 + 10 Sec increment and try it under Arena or Chess Gui at Depth 2 if you are 1500
https://komodochess.com/Komodo14.htm

PS: I spent 4 days watching TCEC Live from Monday thru Thurdays and that game look almost identical to Stockfish Dept Vs Chat except the last 10 moves or so :roll:

lkaufman
Posts: 4286
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: What are Stockfish and Komodo estimated rating for these 5 depths ?

Post by lkaufman » Sat Jul 25, 2020 4:44 am

Milos wrote:
Sat Jul 25, 2020 3:15 am
lkaufman wrote:
Sat Jul 25, 2020 3:04 am
In the 1980s machines that did 4 ply full width (plus qsearch) got ratings like 1600 or so in standard tournaments, and today's 1 ply searches, though "smarter", are nowhere near the level of a 4 ply search with even fairly basic chess knowledge of the mid 1980s.
Again you are just inventing stuff. Not the first time. Care to document your wild claims?
Are you even aware how much more SF extends in qs than those early programs? Just looking at nodes search SFdev 1ply goes over similar number of nodes as full width 4 ply negamax. And not getting into how much more tuned SFdev eval is just for material+PSQT (that btw. didn't even exist in 80s), not to mention king safety, all bonuses, endgame knowledge, etc. etc.
SFdev 1 ply is hundreds of Elo stronger than '80s program 4 ply search.
As far as I know, at one ply stockfish just extends checks, and includes checks on first ply of qsearch, then only nonlosing captures (maybe I'm omitting some other detail, nothing important though). This was all standard in chess engines by at least the late 1980s, as was piece square tables. I ought to know, I was coauthor with Don Dailey of Rexchess, which was maybe about number 5 engine or so in the late 1980s and had these things. I don't think I have any engine now from that period that will run on my modern laptop, but if there is one available that will run, I'll be glad to check this out. I would be surprised if Stockfish or Komodo at one ply could beat Rexchess or another top engine of 1988 or so at two ply; an extra ply makes up for a lot of knowledge at this depth, and anyway all the knowledge in Stockfish and Komodo is tuned for much higher search depths and is quite out of tune for one ply. I think we could easily make Komodo fifty elo stronger or more at one ply if we had a motivation to do so, but it would be weaker at real time controls. I suppose the same is true of Stockfish.
Komodo rules!

Milos
Posts: 3762
Joined: Wed Nov 25, 2009 12:47 am

Re: What are Stockfish and Komodo estimated rating for these 5 depths ?

Post by Milos » Sat Jul 25, 2020 5:41 am

lkaufman wrote:
Sat Jul 25, 2020 4:44 am
As far as I know, at one ply stockfish just extends checks, and includes checks on first ply of qsearch, then only nonlosing captures (maybe I'm omitting some other detail, nothing important though). This was all standard in chess engines by at least the late 1980s, as was piece square tables. I ought to know, I was coauthor with Don Dailey of Rexchess, which was maybe about number 5 engine or so in the late 1980s and had these things. I don't think I have any engine now from that period that will run on my modern laptop, but if there is one available that will run, I'll be glad to check this out. I would be surprised if Stockfish or Komodo at one ply could beat Rexchess or another top engine of 1988 or so at two ply; an extra ply makes up for a lot of knowledge at this depth, and anyway all the knowledge in Stockfish and Komodo is tuned for much higher search depths and is quite out of tune for one ply. I think we could easily make Komodo fifty elo stronger or more at one ply if we had a motivation to do so, but it would be weaker at real time controls. I suppose the same is true of Stockfish.
There are far more details than you obviously aware. First there was no QS in today's sense in Rexchess. Go reread a thread on QS in programming forum. Dan commented it a lot. Most programs in 80's only had SEE and very few QS limited to few plys (i.e. had a depth limit overall not only on check extensions) plus when searching on full window bounds your qsearch stops much, much before than today.
Regarding PSTs anything before Fruit and Rybka when things started to actually be tuned was a joke. As was most of your evaluation concepts. This might have been interesting 30 years ago, but today we know that without systematic tuning it doesn't work at all.

The rest of your post is just speculation, zero evidence about it.

User avatar
Rebel
Posts: 5349
Joined: Thu Aug 18, 2011 10:04 am

Re: What are Stockfish and Komodo estimated rating for these 5 depths ?

Post by Rebel » Sat Jul 25, 2020 6:04 am

Milos wrote:
Sat Jul 25, 2020 5:41 am
lkaufman wrote:
Sat Jul 25, 2020 4:44 am
As far as I know, at one ply stockfish just extends checks, and includes checks on first ply of qsearch, then only nonlosing captures (maybe I'm omitting some other detail, nothing important though). This was all standard in chess engines by at least the late 1980s, as was piece square tables. I ought to know, I was coauthor with Don Dailey of Rexchess, which was maybe about number 5 engine or so in the late 1980s and had these things. I don't think I have any engine now from that period that will run on my modern laptop, but if there is one available that will run, I'll be glad to check this out. I would be surprised if Stockfish or Komodo at one ply could beat Rexchess or another top engine of 1988 or so at two ply; an extra ply makes up for a lot of knowledge at this depth, and anyway all the knowledge in Stockfish and Komodo is tuned for much higher search depths and is quite out of tune for one ply. I think we could easily make Komodo fifty elo stronger or more at one ply if we had a motivation to do so, but it would be weaker at real time controls. I suppose the same is true of Stockfish.
There are far more details than you obviously aware. First there was no QS in today's sense in Rexchess. Go reread a thread on QS in programming forum. Dan commented it a lot. Most programs in 80's only had SEE and very few QS limited to few plys (i.e. had a depth limit overall not only on check extensions) plus when searching on full window bounds your qsearch stops much, much before than today.
Regarding PSTs anything before Fruit and Rybka when things started to actually be tuned was a joke. As was most of your evaluation concepts. This might have been interesting 30 years ago, but today we know that without systematic tuning it doesn't work at all.

The rest of your post is just speculation, zero evidence about it.
I think you are strongly underestimate the oldies, speaking for myself: my evaluation was written without the comfort of eng-eng matches because there was nothing in the 80's. Tuning came from playing manual games at 40/2h, watching and make notes. When eng-eng became available in the mid 90's hardly any evaluation improvement came from that.
90% of coding is debugging, the other 10% is writing bugs.

Milos
Posts: 3762
Joined: Wed Nov 25, 2009 12:47 am

Re: What are Stockfish and Komodo estimated rating for these 5 depths ?

Post by Milos » Sat Jul 25, 2020 6:11 am

Rebel wrote:
Sat Jul 25, 2020 6:04 am
Milos wrote:
Sat Jul 25, 2020 5:41 am
lkaufman wrote:
Sat Jul 25, 2020 4:44 am
As far as I know, at one ply stockfish just extends checks, and includes checks on first ply of qsearch, then only nonlosing captures (maybe I'm omitting some other detail, nothing important though). This was all standard in chess engines by at least the late 1980s, as was piece square tables. I ought to know, I was coauthor with Don Dailey of Rexchess, which was maybe about number 5 engine or so in the late 1980s and had these things. I don't think I have any engine now from that period that will run on my modern laptop, but if there is one available that will run, I'll be glad to check this out. I would be surprised if Stockfish or Komodo at one ply could beat Rexchess or another top engine of 1988 or so at two ply; an extra ply makes up for a lot of knowledge at this depth, and anyway all the knowledge in Stockfish and Komodo is tuned for much higher search depths and is quite out of tune for one ply. I think we could easily make Komodo fifty elo stronger or more at one ply if we had a motivation to do so, but it would be weaker at real time controls. I suppose the same is true of Stockfish.
There are far more details than you obviously aware. First there was no QS in today's sense in Rexchess. Go reread a thread on QS in programming forum. Dan commented it a lot. Most programs in 80's only had SEE and very few QS limited to few plys (i.e. had a depth limit overall not only on check extensions) plus when searching on full window bounds your qsearch stops much, much before than today.
Regarding PSTs anything before Fruit and Rybka when things started to actually be tuned was a joke. As was most of your evaluation concepts. This might have been interesting 30 years ago, but today we know that without systematic tuning it doesn't work at all.

The rest of your post is just speculation, zero evidence about it.
I think you are strongly underestimate the oldies, speaking for myself: my evaluation was written without the comfort of eng-eng matches because there was nothing in the 80's. Tuning came from playing manual games at 40/2h, watching and make notes. When eng-eng became available in the mid 90's hardly any evaluation improvement came from that.
That's exactly why I wrote Fruit and Rybka. It's not about eng-eng matches it's about systematic tuning where one needs sound statistical criteria and possibility to play sufficient number of matches. That was not available before mid 2000's. And that's precisely why first Fruit and than Rybka was such a (quantum) leap.

User avatar
M ANSARI
Posts: 3454
Joined: Thu Mar 16, 2006 6:10 pm

Re: What are Stockfish and Komodo estimated rating for these 5 depths ?

Post by M ANSARI » Sat Jul 25, 2020 6:59 am

That's exactly why I wrote Fruit and Rybka. It's not about eng-eng matches it's about systematic tuning where one needs sound statistical criteria and possibility to play sufficient number of matches. That was not available before mid 2000's. And that's precisely why first Fruit and than Rybka was such a (quantum) leap.


Yes ... and this trend seems to be continuing with the success of SF NN! Would be interesting to see how much stronger at low depth SF is with an NN tuned evaluation.

User avatar
Rebel
Posts: 5349
Joined: Thu Aug 18, 2011 10:04 am

Re: What are Stockfish and Komodo estimated rating for these 5 depths ?

Post by Rebel » Sat Jul 25, 2020 7:07 am

Milos wrote:
Sat Jul 25, 2020 6:11 am
Rebel wrote:
Sat Jul 25, 2020 6:04 am
Milos wrote:
Sat Jul 25, 2020 5:41 am
lkaufman wrote:
Sat Jul 25, 2020 4:44 am
As far as I know, at one ply stockfish just extends checks, and includes checks on first ply of qsearch, then only nonlosing captures (maybe I'm omitting some other detail, nothing important though). This was all standard in chess engines by at least the late 1980s, as was piece square tables. I ought to know, I was coauthor with Don Dailey of Rexchess, which was maybe about number 5 engine or so in the late 1980s and had these things. I don't think I have any engine now from that period that will run on my modern laptop, but if there is one available that will run, I'll be glad to check this out. I would be surprised if Stockfish or Komodo at one ply could beat Rexchess or another top engine of 1988 or so at two ply; an extra ply makes up for a lot of knowledge at this depth, and anyway all the knowledge in Stockfish and Komodo is tuned for much higher search depths and is quite out of tune for one ply. I think we could easily make Komodo fifty elo stronger or more at one ply if we had a motivation to do so, but it would be weaker at real time controls. I suppose the same is true of Stockfish.
There are far more details than you obviously aware. First there was no QS in today's sense in Rexchess. Go reread a thread on QS in programming forum. Dan commented it a lot. Most programs in 80's only had SEE and very few QS limited to few plys (i.e. had a depth limit overall not only on check extensions) plus when searching on full window bounds your qsearch stops much, much before than today.
Regarding PSTs anything before Fruit and Rybka when things started to actually be tuned was a joke. As was most of your evaluation concepts. This might have been interesting 30 years ago, but today we know that without systematic tuning it doesn't work at all.

The rest of your post is just speculation, zero evidence about it.
I think you are strongly underestimate the oldies, speaking for myself: my evaluation was written without the comfort of eng-eng matches because there was nothing in the 80's. Tuning came from playing manual games at 40/2h, watching and make notes. When eng-eng became available in the mid 90's hardly any evaluation improvement came from that.
That's exactly why I wrote Fruit and Rybka. It's not about eng-eng matches it's about systematic tuning where one needs sound statistical criteria and possibility to play sufficient number of matches. That was not available before mid 2000's. And that's precisely why first Fruit and than Rybka was such a (quantum) leap.
You said - Regarding PSTs anything before Fruit and Rybka when things started to actually be tuned was a joke.

You are hurting my feelings :D

Part of the success of Rybka was the realization that volume matters, is crucial and he invested in hardware. In 2005 he wrote in the old CCC -
Handware is important and old ideas work. Back then many working ideas were put aside because too few games were played, also regressions were accpted as improvements for the same reason.

I consider the father of modern testing Christophe Theron, in the late 90's he had just 2 lousy Pentium 90 and he could produce 800 blitz games in 6 hours and it put him on top for a couple of years.
90% of coding is debugging, the other 10% is writing bugs.

Post Reply