Stockfish Handicap Matches

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.
Chessqueen
Posts: 1033
Joined: Wed Sep 05, 2018 12:16 am
Full name: Nancy M Pichardo

Re: Stockfish Handicap Matches

Post by Chessqueen » Fri Jul 10, 2020 5:23 am

lkaufman wrote:
Thu Jul 09, 2020 4:41 pm
Raphexon wrote:
Tue Jul 07, 2020 5:45 pm
lkaufman wrote:
Tue Jul 07, 2020 3:08 pm
Raphexon wrote:
Tue Jul 07, 2020 2:30 pm
lkaufman wrote:
Tue Jul 07, 2020 3:18 am
Chessqueen wrote:
Tue Jul 07, 2020 1:55 am
Vinvin wrote:
Mon Jul 06, 2020 4:32 pm
Chessqueen wrote:
Mon Jul 06, 2020 12:59 pm
Vinvin wrote:
Mon Jul 06, 2020 12:52 am
Vinvin wrote:
Sun Jul 05, 2020 11:55 pm
lkaufman wrote:
Sun Jul 05, 2020 8:38 pm
Vinvin wrote:
Sun Jul 05, 2020 11:55 pm
For the numbers I saw : A 486/66MHz runs around 4 Kn/sec : viewtopic.php?p=821251#p821251
And a recent 4 cores CPU runs around 8 Mn/sec. So, there's no exaggeration for "2000 times".
This is quite remarkable, because I remember clearly, as the operator, that RexChess running on a 486/25 MHz in a WBCA Blitz tournament made an even or better score vs. GMs, including a win over GM Leonid Yudasin, who was a World Champion Candidate just a year later! Based on your numbers, it was at least 5,000 times slower than a recent quad. So if we assume that Komodo (which had the same two original authors as RexChess, although Mark Lefler took over Don Dailey's role in Komodo when he died) is at least not weaker than Rexchess (a fair assumption; probably even on ancient hardware it is much stronger), this suggests that Komodo on a recent quad playing game in 300/5000 = 0.06 seconds would perform at GM level or better in blitz! But that's not plausible; it would be playing about 1 millisecond per move, too fast for valid play. I suppose one error here is that four threads won't work that fast, so we need to set it for game in 0.24 seconds using just one Thread or about 4 ms per move. Even at that speed it's hard to believe it could play even with GMs at blitz, but not impossible. Stockfish can't play decent chess this fast due to PV pruning, so that probably means SF would be quite weak on that 66 MHz hardware at least in blitz.
These considerations remind me an article in French from year 1988 ( http://download.abandonware.org/magazin ... %20032.jpg )
where they say "Botvinnik declared that computers will beat world chess champion when they will reach depth 15 in middle game (classic time control)".
At this time Hitech reaches depth 8 to 9 and Chiptest (predecessor of Deep Thought) reaches depth 9 to 10. EBF value was around 4.
+ The software become so strong and computers become so fast that we can't evaluate anymore how much time does it take for an engine to reach the GM level in classic game.
On the other hand, GM preparation and play has improved a lot (thanks to computers) over the last 10 years also.
I was wondering what would happened if a millionaire pay Carlsen 1 million to play 6 games using the same hardware 486 450 MHz versus the same engine Rebel that played versus Anand at normal time control NOT game in 15 minutes like this https://www.youtube.com/watch?v=BfnOjfP ... 3w&index=2
...
I thought it over for 2 hours and came to the conclusion that even the Rebel that played versus Anand in 2002 on a 486 450 MHz is NOT good enough to beat Carlsen even on game in 90 Minutes, it would have to be one of the current top 5 engines converted to play on a Mere 486 450 MHz
The game you pointed is a long game (2h/40m+1h/all). Game is here : https://lichess.org/study/LMFuhFUn/drLU3NVS
Note that the 486 stopped around 100 MHz. https://en.wikipedia.org/wiki/List_of_I ... processors
450 Mhz was for the Intel Pentium II, Pentium III, AMD-K6-2 or AMD-K6-3 : https://www.rebel.nl/bench.htm
Anand was stronger than Rebel in long games.
For Carlsen, I expect he would win a match 5-1 against the same engine on the same computer.
You are correct I checked the AMD computer that my mother in law gave me, and it is an Old AMD K6-2 450 MHZ Not a 484 450 MHZ, but my question is still the same with any of the current 5 top engines of today if any could be converted to run on my new adquired AMD K6-2 450 MHz can is still beat Carlsen at 90 Minutes per game ? I believe that if Komodo could be compile, it would beat Carlsen 4 out of 6 or probably 4.5 out 6 games.
We know that the best engines of 2002 on the best hardware broke even in matches with Kasparov and Kramnik then. I don't know the speed ratio between the best machine of 2002 and your machine, but it probably isn't a big number, especially since they weren't getting much benefit back then out of using 4 threads. I'm pretty sure that the elo difference between Komodo and any engine of 2002 is vastly greater than the elo gap due to the cheaper hardware, way more than any elo diff. between Carlsen and the two Ks. So I would say your estimate is reasonable or perhaps even conservative.
https://www.chess.com/news/view/time-od ... 5-0-5-6926

I think it's a very conservative estimate.

Komodo from 4 years ago easily beat a GM with 30-1 time odds on 1 core and no pondering.
I don't know what model CPU you used but I assume it's a Xeon, so not even very good single threaded performance.
It was a then current model laptop. But there were other odds too, White every game, only 3 move opening book, no TBs. I don't know the speed ratio of that hardware to to 2002 hardware proposed. But keep in mind that Carlsen has about 270 elo over Erenburg.
If top human GMs play blitz with 3' + 1" (so maybe about 5 seconds per move for typical game), how many milliseconds per move, strict movetime, would Komodo need on one thread to make 50-50 match? This is something we might actually try, so it's not just a hypothetical question. No ponder of course, but otherwise full resources.
I wouldn't be very surprised if 4 ms is enough.
Below my reasoning:

ChessGenius on a Pentium 100 hz managed a 2795 tournament rating (including beating Kasparov) in Rapid (25m +10s)
Since humans are much worse at Blitz than at Rapid, we can assume it would have been able to get a 2795 in Blitz (3m+1s)
SF6 on a Pentium 75hz managed about 6000 nps.
So probably about 8knps on a 100hz.
Modern CPU should get you 1.5mnps with an engine like K or SF, so roughly 187.5 times as fast as the old CPU.
So chessgenius on a modern CPU should be able to get its old rating in blitz with 1s + 5ms so roughly 27ms per move.

Then it's not that big of a leap of faith that Komodo can beat ChessGenius with 7:1 time odds, even at super blitz.

The only problem is that there's some overhead at very low depth so Komodo won't be at full nps until a few ply in.
But 4ms should get Komodo 2000 nodes at start, which is I think enough in Blitz.
I did some tests which shed light on the question. I tested to see how many ms movetime made an even match vs. Komodo skill levels 22 and 23, since we have some human data on them. Skill 22 was equal to 11 ms, Skill 23 was between 18 and 19 ms. In human blitz games, Skill 22 was a bit too strong for GM Alex Lenderman, 21 was more balanced, while in slightly longer blitz games Skill 23 beat most strong GM, but Hikaru did well. It obviously depends on the exact time limit, but it looks like 10 ms is a good match in blitz for a 2600 GM, while 20 ms might be needed vs. Hikaru or Magnus.

Larry If you ever decides to convert Socrate from DOS to work with Windows and for either 32 bits or 64 bits, do you believe that Socrate on a mere AMD K6-2- 450 MHz or early Pentium can beat Carlsen on standard Time control ? https://www.chessprogramming.org/Socrates

lkaufman
Posts: 4289
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Stockfish Handicap Matches

Post by lkaufman » Fri Jul 10, 2020 5:31 am

Chessqueen wrote:
Fri Jul 10, 2020 5:23 am
lkaufman wrote:
Thu Jul 09, 2020 4:41 pm
Raphexon wrote:
Tue Jul 07, 2020 5:45 pm
lkaufman wrote:
Tue Jul 07, 2020 3:08 pm
Raphexon wrote:
Tue Jul 07, 2020 2:30 pm
lkaufman wrote:
Tue Jul 07, 2020 3:18 am
Chessqueen wrote:
Tue Jul 07, 2020 1:55 am
Vinvin wrote:
Mon Jul 06, 2020 4:32 pm
Chessqueen wrote:
Mon Jul 06, 2020 12:59 pm
Vinvin wrote:
Mon Jul 06, 2020 12:52 am
Vinvin wrote:
Sun Jul 05, 2020 11:55 pm
lkaufman wrote:
Sun Jul 05, 2020 8:38 pm
Vinvin wrote:
Sun Jul 05, 2020 11:55 pm
For the numbers I saw : A 486/66MHz runs around 4 Kn/sec : viewtopic.php?p=821251#p821251
And a recent 4 cores CPU runs around 8 Mn/sec. So, there's no exaggeration for "2000 times".
This is quite remarkable, because I remember clearly, as the operator, that RexChess running on a 486/25 MHz in a WBCA Blitz tournament made an even or better score vs. GMs, including a win over GM Leonid Yudasin, who was a World Champion Candidate just a year later! Based on your numbers, it was at least 5,000 times slower than a recent quad. So if we assume that Komodo (which had the same two original authors as RexChess, although Mark Lefler took over Don Dailey's role in Komodo when he died) is at least not weaker than Rexchess (a fair assumption; probably even on ancient hardware it is much stronger), this suggests that Komodo on a recent quad playing game in 300/5000 = 0.06 seconds would perform at GM level or better in blitz! But that's not plausible; it would be playing about 1 millisecond per move, too fast for valid play. I suppose one error here is that four threads won't work that fast, so we need to set it for game in 0.24 seconds using just one Thread or about 4 ms per move. Even at that speed it's hard to believe it could play even with GMs at blitz, but not impossible. Stockfish can't play decent chess this fast due to PV pruning, so that probably means SF would be quite weak on that 66 MHz hardware at least in blitz.
These considerations remind me an article in French from year 1988 ( http://download.abandonware.org/magazin ... %20032.jpg )
where they say "Botvinnik declared that computers will beat world chess champion when they will reach depth 15 in middle game (classic time control)".
At this time Hitech reaches depth 8 to 9 and Chiptest (predecessor of Deep Thought) reaches depth 9 to 10. EBF value was around 4.
+ The software become so strong and computers become so fast that we can't evaluate anymore how much time does it take for an engine to reach the GM level in classic game.
On the other hand, GM preparation and play has improved a lot (thanks to computers) over the last 10 years also.
I was wondering what would happened if a millionaire pay Carlsen 1 million to play 6 games using the same hardware 486 450 MHz versus the same engine Rebel that played versus Anand at normal time control NOT game in 15 minutes like this https://www.youtube.com/watch?v=BfnOjfP ... 3w&index=2
...
I thought it over for 2 hours and came to the conclusion that even the Rebel that played versus Anand in 2002 on a 486 450 MHz is NOT good enough to beat Carlsen even on game in 90 Minutes, it would have to be one of the current top 5 engines converted to play on a Mere 486 450 MHz
The game you pointed is a long game (2h/40m+1h/all). Game is here : https://lichess.org/study/LMFuhFUn/drLU3NVS
Note that the 486 stopped around 100 MHz. https://en.wikipedia.org/wiki/List_of_I ... processors
450 Mhz was for the Intel Pentium II, Pentium III, AMD-K6-2 or AMD-K6-3 : https://www.rebel.nl/bench.htm
Anand was stronger than Rebel in long games.
For Carlsen, I expect he would win a match 5-1 against the same engine on the same computer.
You are correct I checked the AMD computer that my mother in law gave me, and it is an Old AMD K6-2 450 MHZ Not a 484 450 MHZ, but my question is still the same with any of the current 5 top engines of today if any could be converted to run on my new adquired AMD K6-2 450 MHz can is still beat Carlsen at 90 Minutes per game ? I believe that if Komodo could be compile, it would beat Carlsen 4 out of 6 or probably 4.5 out 6 games.
We know that the best engines of 2002 on the best hardware broke even in matches with Kasparov and Kramnik then. I don't know the speed ratio between the best machine of 2002 and your machine, but it probably isn't a big number, especially since they weren't getting much benefit back then out of using 4 threads. I'm pretty sure that the elo difference between Komodo and any engine of 2002 is vastly greater than the elo gap due to the cheaper hardware, way more than any elo diff. between Carlsen and the two Ks. So I would say your estimate is reasonable or perhaps even conservative.
https://www.chess.com/news/view/time-od ... 5-0-5-6926

I think it's a very conservative estimate.

Komodo from 4 years ago easily beat a GM with 30-1 time odds on 1 core and no pondering.
I don't know what model CPU you used but I assume it's a Xeon, so not even very good single threaded performance.
It was a then current model laptop. But there were other odds too, White every game, only 3 move opening book, no TBs. I don't know the speed ratio of that hardware to to 2002 hardware proposed. But keep in mind that Carlsen has about 270 elo over Erenburg.
If top human GMs play blitz with 3' + 1" (so maybe about 5 seconds per move for typical game), how many milliseconds per move, strict movetime, would Komodo need on one thread to make 50-50 match? This is something we might actually try, so it's not just a hypothetical question. No ponder of course, but otherwise full resources.
I wouldn't be very surprised if 4 ms is enough.
Below my reasoning:

ChessGenius on a Pentium 100 hz managed a 2795 tournament rating (including beating Kasparov) in Rapid (25m +10s)
Since humans are much worse at Blitz than at Rapid, we can assume it would have been able to get a 2795 in Blitz (3m+1s)
SF6 on a Pentium 75hz managed about 6000 nps.
So probably about 8knps on a 100hz.
Modern CPU should get you 1.5mnps with an engine like K or SF, so roughly 187.5 times as fast as the old CPU.
So chessgenius on a modern CPU should be able to get its old rating in blitz with 1s + 5ms so roughly 27ms per move.

Then it's not that big of a leap of faith that Komodo can beat ChessGenius with 7:1 time odds, even at super blitz.

The only problem is that there's some overhead at very low depth so Komodo won't be at full nps until a few ply in.
But 4ms should get Komodo 2000 nodes at start, which is I think enough in Blitz.
I did some tests which shed light on the question. I tested to see how many ms movetime made an even match vs. Komodo skill levels 22 and 23, since we have some human data on them. Skill 22 was equal to 11 ms, Skill 23 was between 18 and 19 ms. In human blitz games, Skill 22 was a bit too strong for GM Alex Lenderman, 21 was more balanced, while in slightly longer blitz games Skill 23 beat most strong GM, but Hikaru did well. It obviously depends on the exact time limit, but it looks like 10 ms is a good match in blitz for a 2600 GM, while 20 ms might be needed vs. Hikaru or Magnus.

Larry If you ever decides to convert Socrate from DOS to work with Windows and for either 32 bits or 64 bits, do you believe that Socrate on a mere AMD K6-2- 450 MHz or early Pentium can beat Carlsen on standard Time control ? https://www.chessprogramming.org/Socrates
No. If the best software on the best hardware in 2002 were just even with Kasparov and Kramnik at standard time control, why would you expect software from 1992 running on inferior hardware to beat Carlsen? The progress from Socrates in 1992 to Komodo in 2020 is just enormous, perhaps even more than the hardware has advanced.
Komodo rules!

Chessqueen
Posts: 1033
Joined: Wed Sep 05, 2018 12:16 am
Full name: Nancy M Pichardo

Re: Stockfish Handicap Matches

Post by Chessqueen » Fri Jul 10, 2020 10:18 am

lkaufman wrote:
Fri Jul 10, 2020 5:31 am
Chessqueen wrote:
Fri Jul 10, 2020 5:23 am
lkaufman wrote:
Thu Jul 09, 2020 4:41 pm
Raphexon wrote:
Tue Jul 07, 2020 5:45 pm
lkaufman wrote:
Tue Jul 07, 2020 3:08 pm
Raphexon wrote:
Tue Jul 07, 2020 2:30 pm
lkaufman wrote:
Tue Jul 07, 2020 3:18 am
Chessqueen wrote:
Tue Jul 07, 2020 1:55 am
Vinvin wrote:
Mon Jul 06, 2020 4:32 pm
Chessqueen wrote:
Mon Jul 06, 2020 12:59 pm
Vinvin wrote:
Mon Jul 06, 2020 12:52 am
Vinvin wrote:
Sun Jul 05, 2020 11:55 pm
lkaufman wrote:
Sun Jul 05, 2020 8:38 pm
Vinvin wrote:
Sun Jul 05, 2020 11:55 pm
For the numbers I saw : A 486/66MHz runs around 4 Kn/sec : viewtopic.php?p=821251#p821251
And a recent 4 cores CPU runs around 8 Mn/sec. So, there's no exaggeration for "2000 times".
This is quite remarkable, because I remember clearly, as the operator, that RexChess running on a 486/25 MHz in a WBCA Blitz tournament made an even or better score vs. GMs, including a win over GM Leonid Yudasin, who was a World Champion Candidate just a year later! Based on your numbers, it was at least 5,000 times slower than a recent quad. So if we assume that Komodo (which had the same two original authors as RexChess, although Mark Lefler took over Don Dailey's role in Komodo when he died) is at least not weaker than Rexchess (a fair assumption; probably even on ancient hardware it is much stronger), this suggests that Komodo on a recent quad playing game in 300/5000 = 0.06 seconds would perform at GM level or better in blitz! But that's not plausible; it would be playing about 1 millisecond per move, too fast for valid play. I suppose one error here is that four threads won't work that fast, so we need to set it for game in 0.24 seconds using just one Thread or about 4 ms per move. Even at that speed it's hard to believe it could play even with GMs at blitz, but not impossible. Stockfish can't play decent chess this fast due to PV pruning, so that probably means SF would be quite weak on that 66 MHz hardware at least in blitz.
These considerations remind me an article in French from year 1988 ( http://download.abandonware.org/magazin ... %20032.jpg )
where they say "Botvinnik declared that computers will beat world chess champion when they will reach depth 15 in middle game (classic time control)".
At this time Hitech reaches depth 8 to 9 and Chiptest (predecessor of Deep Thought) reaches depth 9 to 10. EBF value was around 4.
+ The software become so strong and computers become so fast that we can't evaluate anymore how much time does it take for an engine to reach the GM level in classic game.
On the other hand, GM preparation and play has improved a lot (thanks to computers) over the last 10 years also.
I was wondering what would happened if a millionaire pay Carlsen 1 million to play 6 games using the same hardware 486 450 MHz versus the same engine Rebel that played versus Anand at normal time control NOT game in 15 minutes like this https://www.youtube.com/watch?v=BfnOjfP ... 3w&index=2
...
I thought it over for 2 hours and came to the conclusion that even the Rebel that played versus Anand in 2002 on a 486 450 MHz is NOT good enough to beat Carlsen even on game in 90 Minutes, it would have to be one of the current top 5 engines converted to play on a Mere 486 450 MHz
The game you pointed is a long game (2h/40m+1h/all). Game is here : https://lichess.org/study/LMFuhFUn/drLU3NVS
Note that the 486 stopped around 100 MHz. https://en.wikipedia.org/wiki/List_of_I ... processors
450 Mhz was for the Intel Pentium II, Pentium III, AMD-K6-2 or AMD-K6-3 : https://www.rebel.nl/bench.htm
Anand was stronger than Rebel in long games.
For Carlsen, I expect he would win a match 5-1 against the same engine on the same computer.
You are correct I checked the AMD computer that my mother in law gave me, and it is an Old AMD K6-2 450 MHZ Not a 484 450 MHZ, but my question is still the same with any of the current 5 top engines of today if any could be converted to run on my new adquired AMD K6-2 450 MHz can is still beat Carlsen at 90 Minutes per game ? I believe that if Komodo could be compile, it would beat Carlsen 4 out of 6 or probably 4.5 out 6 games.
We know that the best engines of 2002 on the best hardware broke even in matches with Kasparov and Kramnik then. I don't know the speed ratio between the best machine of 2002 and your machine, but it probably isn't a big number, especially since they weren't getting much benefit back then out of using 4 threads. I'm pretty sure that the elo difference between Komodo and any engine of 2002 is vastly greater than the elo gap due to the cheaper hardware, way more than any elo diff. between Carlsen and the two Ks. So I would say your estimate is reasonable or perhaps even conservative.
https://www.chess.com/news/view/time-od ... 5-0-5-6926

I think it's a very conservative estimate.

Komodo from 4 years ago easily beat a GM with 30-1 time odds on 1 core and no pondering.
I don't know what model CPU you used but I assume it's a Xeon, so not even very good single threaded performance.
It was a then current model laptop. But there were other odds too, White every game, only 3 move opening book, no TBs. I don't know the speed ratio of that hardware to to 2002 hardware proposed. But keep in mind that Carlsen has about 270 elo over Erenburg.
If top human GMs play blitz with 3' + 1" (so maybe about 5 seconds per move for typical game), how many milliseconds per move, strict movetime, would Komodo need on one thread to make 50-50 match? This is something we might actually try, so it's not just a hypothetical question. No ponder of course, but otherwise full resources.
I wouldn't be very surprised if 4 ms is enough.
Below my reasoning:

ChessGenius on a Pentium 100 hz managed a 2795 tournament rating (including beating Kasparov) in Rapid (25m +10s)
Since humans are much worse at Blitz than at Rapid, we can assume it would have been able to get a 2795 in Blitz (3m+1s)
SF6 on a Pentium 75hz managed about 6000 nps.
So probably about 8knps on a 100hz.
Modern CPU should get you 1.5mnps with an engine like K or SF, so roughly 187.5 times as fast as the old CPU.
So chessgenius on a modern CPU should be able to get its old rating in blitz with 1s + 5ms so roughly 27ms per move.

Then it's not that big of a leap of faith that Komodo can beat ChessGenius with 7:1 time odds, even at super blitz.

The only problem is that there's some overhead at very low depth so Komodo won't be at full nps until a few ply in.
But 4ms should get Komodo 2000 nodes at start, which is I think enough in Blitz.
I did some tests which shed light on the question. I tested to see how many ms movetime made an even match vs. Komodo skill levels 22 and 23, since we have some human data on them. Skill 22 was equal to 11 ms, Skill 23 was between 18 and 19 ms. In human blitz games, Skill 22 was a bit too strong for GM Alex Lenderman, 21 was more balanced, while in slightly longer blitz games Skill 23 beat most strong GM, but Hikaru did well. It obviously depends on the exact time limit, but it looks like 10 ms is a good match in blitz for a 2600 GM, while 20 ms might be needed vs. Hikaru or Magnus.

Larry If you ever decides to convert Socrate from DOS to work with Windows and for either 32 bits or 64 bits, do you believe that Socrate on a mere AMD K6-2- 450 MHz or early Pentium can beat Carlsen on standard Time control ? https://www.chessprogramming.org/Socrates
No. If the best software on the best hardware in 2002 were just even with Kasparov and Kramnik at standard time control, why would you expect software from 1992 running on inferior hardware to beat Carlsen? The progress from Socrates in 1992 to Komodo in 2020 is just enormous, perhaps even more than the hardware has advanced.
Okay I will use DOS-BOX to run Socrate under windows and match it versus Jorge Sammour using my AMD K6-2 450 MHz. Where can I find Socrate can you make it available? https://www.youtube.com/watch?v=5yBr-n_ee2E

PS: Here is DOSBOX ==> https://www.dosbox.com/

Vinvin
Posts: 4690
Joined: Thu Mar 09, 2006 8:40 am
Full name: Vincent Lejeune

Re: Stockfish Handicap Matches

Post by Vinvin » Fri Jul 10, 2020 10:29 am

Chessqueen wrote:
Fri Jul 10, 2020 10:18 am
lkaufman wrote:
Fri Jul 10, 2020 5:31 am
Chessqueen wrote:
Fri Jul 10, 2020 5:23 am
lkaufman wrote:
Thu Jul 09, 2020 4:41 pm
Raphexon wrote:
Tue Jul 07, 2020 5:45 pm
lkaufman wrote:
Tue Jul 07, 2020 3:08 pm
Raphexon wrote:
Tue Jul 07, 2020 2:30 pm
lkaufman wrote:
Tue Jul 07, 2020 3:18 am
Chessqueen wrote:
Tue Jul 07, 2020 1:55 am
Vinvin wrote:
Mon Jul 06, 2020 4:32 pm
Chessqueen wrote:
Mon Jul 06, 2020 12:59 pm
Vinvin wrote:
Mon Jul 06, 2020 12:52 am
Vinvin wrote:
Sun Jul 05, 2020 11:55 pm
lkaufman wrote:
Sun Jul 05, 2020 8:38 pm
Vinvin wrote:
Sun Jul 05, 2020 11:55 pm
For the numbers I saw : A 486/66MHz runs around 4 Kn/sec : viewtopic.php?p=821251#p821251
And a recent 4 cores CPU runs around 8 Mn/sec. So, there's no exaggeration for "2000 times".
This is quite remarkable, because I remember clearly, as the operator, that RexChess running on a 486/25 MHz in a WBCA Blitz tournament made an even or better score vs. GMs, including a win over GM Leonid Yudasin, who was a World Champion Candidate just a year later! Based on your numbers, it was at least 5,000 times slower than a recent quad. So if we assume that Komodo (which had the same two original authors as RexChess, although Mark Lefler took over Don Dailey's role in Komodo when he died) is at least not weaker than Rexchess (a fair assumption; probably even on ancient hardware it is much stronger), this suggests that Komodo on a recent quad playing game in 300/5000 = 0.06 seconds would perform at GM level or better in blitz! But that's not plausible; it would be playing about 1 millisecond per move, too fast for valid play. I suppose one error here is that four threads won't work that fast, so we need to set it for game in 0.24 seconds using just one Thread or about 4 ms per move. Even at that speed it's hard to believe it could play even with GMs at blitz, but not impossible. Stockfish can't play decent chess this fast due to PV pruning, so that probably means SF would be quite weak on that 66 MHz hardware at least in blitz.
These considerations remind me an article in French from year 1988 ( http://download.abandonware.org/magazin ... %20032.jpg )
where they say "Botvinnik declared that computers will beat world chess champion when they will reach depth 15 in middle game (classic time control)".
At this time Hitech reaches depth 8 to 9 and Chiptest (predecessor of Deep Thought) reaches depth 9 to 10. EBF value was around 4.
+ The software become so strong and computers become so fast that we can't evaluate anymore how much time does it take for an engine to reach the GM level in classic game.
On the other hand, GM preparation and play has improved a lot (thanks to computers) over the last 10 years also.
I was wondering what would happened if a millionaire pay Carlsen 1 million to play 6 games using the same hardware 486 450 MHz versus the same engine Rebel that played versus Anand at normal time control NOT game in 15 minutes like this https://www.youtube.com/watch?v=BfnOjfP ... 3w&index=2
...
I thought it over for 2 hours and came to the conclusion that even the Rebel that played versus Anand in 2002 on a 486 450 MHz is NOT good enough to beat Carlsen even on game in 90 Minutes, it would have to be one of the current top 5 engines converted to play on a Mere 486 450 MHz
The game you pointed is a long game (2h/40m+1h/all). Game is here : https://lichess.org/study/LMFuhFUn/drLU3NVS
Note that the 486 stopped around 100 MHz. https://en.wikipedia.org/wiki/List_of_I ... processors
450 Mhz was for the Intel Pentium II, Pentium III, AMD-K6-2 or AMD-K6-3 : https://www.rebel.nl/bench.htm
Anand was stronger than Rebel in long games.
For Carlsen, I expect he would win a match 5-1 against the same engine on the same computer.
You are correct I checked the AMD computer that my mother in law gave me, and it is an Old AMD K6-2 450 MHZ Not a 484 450 MHZ, but my question is still the same with any of the current 5 top engines of today if any could be converted to run on my new adquired AMD K6-2 450 MHz can is still beat Carlsen at 90 Minutes per game ? I believe that if Komodo could be compile, it would beat Carlsen 4 out of 6 or probably 4.5 out 6 games.
We know that the best engines of 2002 on the best hardware broke even in matches with Kasparov and Kramnik then. I don't know the speed ratio between the best machine of 2002 and your machine, but it probably isn't a big number, especially since they weren't getting much benefit back then out of using 4 threads. I'm pretty sure that the elo difference between Komodo and any engine of 2002 is vastly greater than the elo gap due to the cheaper hardware, way more than any elo diff. between Carlsen and the two Ks. So I would say your estimate is reasonable or perhaps even conservative.
https://www.chess.com/news/view/time-od ... 5-0-5-6926

I think it's a very conservative estimate.

Komodo from 4 years ago easily beat a GM with 30-1 time odds on 1 core and no pondering.
I don't know what model CPU you used but I assume it's a Xeon, so not even very good single threaded performance.
It was a then current model laptop. But there were other odds too, White every game, only 3 move opening book, no TBs. I don't know the speed ratio of that hardware to to 2002 hardware proposed. But keep in mind that Carlsen has about 270 elo over Erenburg.
If top human GMs play blitz with 3' + 1" (so maybe about 5 seconds per move for typical game), how many milliseconds per move, strict movetime, would Komodo need on one thread to make 50-50 match? This is something we might actually try, so it's not just a hypothetical question. No ponder of course, but otherwise full resources.
I wouldn't be very surprised if 4 ms is enough.
Below my reasoning:

ChessGenius on a Pentium 100 hz managed a 2795 tournament rating (including beating Kasparov) in Rapid (25m +10s)
Since humans are much worse at Blitz than at Rapid, we can assume it would have been able to get a 2795 in Blitz (3m+1s)
SF6 on a Pentium 75hz managed about 6000 nps.
So probably about 8knps on a 100hz.
Modern CPU should get you 1.5mnps with an engine like K or SF, so roughly 187.5 times as fast as the old CPU.
So chessgenius on a modern CPU should be able to get its old rating in blitz with 1s + 5ms so roughly 27ms per move.

Then it's not that big of a leap of faith that Komodo can beat ChessGenius with 7:1 time odds, even at super blitz.

The only problem is that there's some overhead at very low depth so Komodo won't be at full nps until a few ply in.
But 4ms should get Komodo 2000 nodes at start, which is I think enough in Blitz.
I did some tests which shed light on the question. I tested to see how many ms movetime made an even match vs. Komodo skill levels 22 and 23, since we have some human data on them. Skill 22 was equal to 11 ms, Skill 23 was between 18 and 19 ms. In human blitz games, Skill 22 was a bit too strong for GM Alex Lenderman, 21 was more balanced, while in slightly longer blitz games Skill 23 beat most strong GM, but Hikaru did well. It obviously depends on the exact time limit, but it looks like 10 ms is a good match in blitz for a 2600 GM, while 20 ms might be needed vs. Hikaru or Magnus.

Larry If you ever decides to convert Socrate from DOS to work with Windows and for either 32 bits or 64 bits, do you believe that Socrate on a mere AMD K6-2- 450 MHz or early Pentium can beat Carlsen on standard Time control ? https://www.chessprogramming.org/Socrates
No. If the best software on the best hardware in 2002 were just even with Kasparov and Kramnik at standard time control, why would you expect software from 1992 running on inferior hardware to beat Carlsen? The progress from Socrates in 1992 to Komodo in 2020 is just enormous, perhaps even more than the hardware has advanced.
Okay I will use DOS-BOX to run Socrate under windows and match it versus Jorge Sammour using my AMD K6-2 450 MHz. Where can I find Socrate can you make it available? https://www.youtube.com/watch?v=5yBr-n_ee2E

PS: Here is DOSBOX ==> https://www.dosbox.com/
Why do you use DOS-BOX ? It will decrease the speed of the program by a factor 20 ...

User avatar
Rebel
Posts: 5354
Joined: Thu Aug 18, 2011 10:04 am

Re: Stockfish Handicap Matches

Post by Rebel » Fri Jul 10, 2020 10:31 am

Chessqueen wrote:
Fri Jul 10, 2020 10:18 am
Okay I will use DOS-BOX to run Socrate under windows and match it versus Jorge Sammour using my AMD K6-2 450 MHz. Where can I find Socrate can you make it available? https://www.youtube.com/watch?v=5yBr-n_ee2E
Socrates and others -
http://rebel13.nl/noomen/more_oldies.html
http://rebel13.nl/noomen/more-dos-oldies.html
90% of coding is debugging, the other 10% is writing bugs.

JJJ
Posts: 1314
Joined: Sat Apr 19, 2014 11:47 am

Re: Stockfish Handicap Matches

Post by JJJ » Fri Jul 10, 2020 3:36 pm

Larry, if I want to play against Komodo at rook handicap at chess.com, is it possible ?
There is many skill level for Komodo, but is there many handicap level to play against ?

We could have an accurate rating of Komodo based on any handicap like this.

Chessqueen
Posts: 1033
Joined: Wed Sep 05, 2018 12:16 am
Full name: Nancy M Pichardo

Re: Stockfish Handicap Matches

Post by Chessqueen » Fri Jul 10, 2020 7:59 pm

JJJ wrote:
Fri Jul 10, 2020 3:36 pm
Larry, if I want to play against Komodo at rook handicap at chess.com, is it possible ?
There is many skill level for Komodo, but is there many handicap level to play against ?

We could have an accurate rating of Komodo based on any handicap like this.
I believe that if lots of people with different FIDE ratings play against Komodo at Odds on Chess.com we can come with a good estimate of different Komodo Odds ratings comparable human ratings given to different players rated from 1600 up to 2750 depending on their FIDE ratings. Probably Komodo should start offering one weeks with a Rook Odds to players between 1600 to 1800, then the next week Rook for let say the b7, or c7 pawn for players rated between 1800 to 2250, then the Next week Komodo would be available to players between 2250 to 2400 with Knight odds, and the next week Komodo could be set to offer any players on chess.com Knight odds and b7 or c7 removed for the human FIDE rated between 2400 to 2675. In other word every week Komodo would offer different Odds to human players of different FIDE ratings.

lkaufman
Posts: 4289
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Stockfish Handicap Matches

Post by lkaufman » Fri Jul 10, 2020 9:56 pm

Rebel wrote:
Fri Jul 10, 2020 10:31 am
Chessqueen wrote:
Fri Jul 10, 2020 10:18 am
Okay I will use DOS-BOX to run Socrate under windows and match it versus Jorge Sammour using my AMD K6-2 450 MHz. Where can I find Socrate can you make it available? https://www.youtube.com/watch?v=5yBr-n_ee2E
Socrates and others -
http://rebel13.nl/noomen/more_oldies.html
http://rebel13.nl/noomen/more-dos-oldies.html
After nearly thirty years, I don't have a copy available, perhaps someone else does. The commercial version was known as "Kasparov's Gambit", perhaps someone has that. Or even the earlier "RexChess".
Komodo rules!

lkaufman
Posts: 4289
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Stockfish Handicap Matches

Post by lkaufman » Fri Jul 10, 2020 10:07 pm

JJJ wrote:
Fri Jul 10, 2020 3:36 pm
Larry, if I want to play against Komodo at rook handicap at chess.com, is it possible ?
There is many skill level for Komodo, but is there many handicap level to play against ?

We could have an accurate rating of Komodo based on any handicap like this.
Chess.com does not have an automatic way for people to play handicap games, either with each other or with computer. I can arrange to have Komodo play against any individual at any handicap (as long as it just involves a specified start position), but we have to agree on the conditions and the starting time for the game or games. It would use my 32 core threadripper (unless we agree to limit the number of threads used for some reason). In theory I could have the computer online and available at a specified time, like every Wednesday night at 7 pm easterm time for example, and select one opponent after another from those who messaged me that they would like to be next. We should probably standardize a time limit to be the same for everyone. What is your own rating or ratings (FIDE, national, online rapid/blitz)? That helps to judge what is reasonable. As the handicap gets larger, the play of any engine gets weaker relative to what Carlsen or Naka could do, because objective merit of movesbecomes less relevant than guessing what will fool the opponent. Probably somewhere between giving knight and rook odds the engine will no longer outperform the best human for this reason. But it will still beat most people.
Komodo rules!

User avatar
Rebel
Posts: 5354
Joined: Thu Aug 18, 2011 10:04 am

Re: Stockfish Handicap Matches

Post by Rebel » Sat Jul 11, 2020 7:36 am

lkaufman wrote:
Fri Jul 10, 2020 9:56 pm
Rebel wrote:
Fri Jul 10, 2020 10:31 am
Chessqueen wrote:
Fri Jul 10, 2020 10:18 am
Okay I will use DOS-BOX to run Socrate under windows and match it versus Jorge Sammour using my AMD K6-2 450 MHz. Where can I find Socrate can you make it available? https://www.youtube.com/watch?v=5yBr-n_ee2E
Socrates and others -
http://rebel13.nl/noomen/more_oldies.html
http://rebel13.nl/noomen/more-dos-oldies.html
After nearly thirty years, I don't have a copy available, perhaps someone else does. The commercial version was known as "Kasparov's Gambit", perhaps someone has that. Or even the earlier "RexChess".
Both are in the the second link above.
90% of coding is debugging, the other 10% is writing bugs.

Post Reply