Komodo vs. Movsesian

[Ozymandias]

He won't succeed in closing everything, if playing against latest Komodo with high contempt set.

But that only works against weaker opponents, and the whole point of time odds, is to equalize the playing field. The GM might try to bust the engine, but only if the prize conditions reward that bet.

[lkaufman]

He won't succeed in closing everything, if playing against latest Komodo with high contempt set.

But that only works against weaker opponents, and the whole point of time odds, is to equalize the playing field. The GM might try to bust the engine, but only if the prize conditions reward that bet.

Even if Komodo were crippled to the point of having only equal chances with a GM opponent, I would want to set it to go for open positions, just because of the different strengths of humans and computers. Humans may have learned to try to close the position in the last decade, but at least some engines have learned to try to open them against humans, so these effects may cancel out.

[thekingman]

I am highly optimistic about Komodo's chances in the pawn and move games. The material advantage is too small for the human to be able to count on simply trading into an advantageous endgame. This will force active play, which gives Komodo strong chances.

[lkaufman]

I am highly optimistic about Komodo's chances in the pawn and move games. The material advantage is too small for the human to be able to count on simply trading into an advantageous endgame. This will force active play, which gives Komodo strong chances.

Yes, to some extent I agree. I've analyzed the f7 handicap quite a bit now in preparation for the match, and while in every variation White gets a substantial positional advantage plus an extra pawn, at least the positions are mostly fairly interesting, not simple endgames or mating attacks. There is enough choice in defense options to avoid any repeat losses at least. But with a very good position and an extra pawn, White really shouldn't lose, and may win some of the games. Movsesian will be well-prepared, I'm sure.

[Ozymandias]

Even if Komodo were crippled to the point of having only equal chances with a GM opponent, I would want to set it to go for open positions, just because of the different strengths of humans and computers. Humans may have learned to try to close the position in the last decade, but at least some engines have learned to try to open them against humans, so these effects may cancel out.

That's the kind of fight, I'm talking about.

[Laskos]

who cares if Komodo plays 15 seconds/game or 40 seconds/game?

I do.

This quantity is not universal, it's very hardware dependent. On 24 cores it will boil down whether to use 4 seconds per game or 5 seconds per game, strength is very sensitive to time control and hardware at these short times. Will depend on overhead, clock tick, parallelization for ultrafast. Isn't it ridiculous to use 24 cores only in order to decrease further the time control of Komodo to the shortest possible values? Just a bit more reasonable would be to use CPU throttling software on a core of a notebook, decreasing one core CPU speed to say 1%, or 10 kN/s for Komodo, the level of 1993 Intel 486DX 50 MHz machine. On this hardware, top 1993 engines were below FM level, and this match would show purely software achievement to 2015, beating on thus emulated machine a strong GM. I find this CPU process throttler useful:
http://mion.faireal.net/BES/
Maybe I will let play overnight at long time control, say 40 moves in 60 minutes the throttled Komodo 9.2 to 10 kN/s against 1 full core Shredder Paderborn, which was top GM level in 2002. Still, not very enlightening. Larry's material handicaps are more illustrative.

[Ozymandias]

who cares if Komodo plays 15 seconds/game or 40 seconds/game?

I do.

This quantity is not universal, it's very hardware dependent. On 24 cores it will boil down whether to use 4 seconds per game or 5 seconds per game, strength is very sensitive to time control and hardware at these short times. Will depend on overhead, clock tick, parallelization for ultrafast. Isn't it ridiculous to use 24 cores only in order to decrease further the time control of Komodo to the shortest possible values? Just a bit more reasonable would be to use CPU throttling software on a core of a notebook, decreasing one core CPU speed to say 1%, or 10 kN/s for Komodo, the level of 1993 Intel 486DX 50 MHz machine. On this hardware, top 1993 engines were below FM level, and this match would show purely software achievement to 2015, beating on thus emulated machine a strong GM. I find this CPU process throttler useful:
http://mion.faireal.net/BES/
Maybe I will let play overnight at long time control, say 40 moves in 60 minutes the throttled Komodo 9.2 to 10 kN/s against 1 full core Shredder Paderborn, which was top GM level in 2002. Still, not very enlightening. Larry's material handicaps are more illustrative.

It may seem silly, but it allows 99% of the game time, to be allocated to the GM. Under the same match conditions, he gets to use more time for his moves, instead of pondering. The GM's time is more expensive, so it's reasonable to make the most, out of it.
Of course such a match is very much HW dependent, but as long as the system specs are published, anyone interested can gauge what it means. The general public, who doesn't care about such details, would be more impressed by the TC, than anything else. Even today, the common chess players, has no idea just how strong engines are, even at blitz/bullet. They all learned years ago, that machines are better at long TC, it's about time they learned a new lesson.

[Laskos]

who cares if Komodo plays 15 seconds/game or 40 seconds/game?

I do.

This quantity is not universal, it's very hardware dependent. On 24 cores it will boil down whether to use 4 seconds per game or 5 seconds per game, strength is very sensitive to time control and hardware at these short times. Will depend on overhead, clock tick, parallelization for ultrafast. Isn't it ridiculous to use 24 cores only in order to decrease further the time control of Komodo to the shortest possible values? Just a bit more reasonable would be to use CPU throttling software on a core of a notebook, decreasing one core CPU speed to say 1%, or 10 kN/s for Komodo, the level of 1993 Intel 486DX 50 MHz machine. On this hardware, top 1993 engines were below FM level, and this match would show purely software achievement to 2015, beating on thus emulated machine a strong GM. I find this CPU process throttler useful:
http://mion.faireal.net/BES/
Maybe I will let play overnight at long time control, say 40 moves in 60 minutes the throttled Komodo 9.2 to 10 kN/s against 1 full core Shredder Paderborn, which was top GM level in 2002. Still, not very enlightening. Larry's material handicaps are more illustrative.

It may seem silly, but it allows 99% of the game time, to be allocated to the GM. Under the same match conditions, he gets to use more time for his moves, instead of pondering. The GM's time is more expensive, so it's reasonable to make the most, out of it.
Of course such a match is very much HW dependent, but as long as the system specs are published, anyone interested can gauge what it means. The general public, who doesn't care about such details, would be more impressed by the TC, than anything else. Even today, the common chess players, has no idea just how strong engines are, even at blitz/bullet. They all learned years ago, that machines are better at long TC, it's about time they learned a new lesson.

I am not sure they do not know that engines are very strong at short time controls, in fact it's likely that most chess players had several games against blitz settings rather than long control. I tested the more reasonable "throttling" of Komodo handicap at long time control (throttled to 10 kN/s or 1993 level 486-50 MHz). As a "human" contender I took Shredder 6 Paderborn, which on full modern core is pretty well calibrated to have a FIDE rating of about 2750 at tournament time control. I used long time control of 90 minutes + 15 seconds for both.



NPS are highlighted for both engines in the picture. The result after 12 games for the throttled Komodo: +6 -2 =4 against full Shredder, so Komodo on 1993 PC hardware is a bit above FIDE 2750. This roughly 100 speed (or time) handicap would be good enough to use against GM Movsesian, still with large chances to beat him at regular chess. If the games are with human at 45'+15'', then Komodo would need 27''+0.15'' on one modern core to be top GM level. But throttling to 1992-93 hardware seems better, because the feeling that the engine moves instantly is probably unsettling. Also, one can compare directly with 1992-93, when on the same hardware top engines were barely weak FM level. And one can separate the improvement during the last 22 years in FIDE ELO: about 500 software, about 400 hardware. All in all, from 2300 to 3200 FIDE ELO at tournament time control.

[lkaufman]

who cares if Komodo plays 15 seconds/game or 40 seconds/game?

I do.

This quantity is not universal, it's very hardware dependent. On 24 cores it will boil down whether to use 4 seconds per game or 5 seconds per game, strength is very sensitive to time control and hardware at these short times. Will depend on overhead, clock tick, parallelization for ultrafast. Isn't it ridiculous to use 24 cores only in order to decrease further the time control of Komodo to the shortest possible values? Just a bit more reasonable would be to use CPU throttling software on a core of a notebook, decreasing one core CPU speed to say 1%, or 10 kN/s for Komodo, the level of 1993 Intel 486DX 50 MHz machine. On this hardware, top 1993 engines were below FM level, and this match would show purely software achievement to 2015, beating on thus emulated machine a strong GM. I find this CPU process throttler useful:
http://mion.faireal.net/BES/
Maybe I will let play overnight at long time control, say 40 moves in 60 minutes the throttled Komodo 9.2 to 10 kN/s against 1 full core Shredder Paderborn, which was top GM level in 2002. Still, not very enlightening. Larry's material handicaps are more illustrative.

It may seem silly, but it allows 99% of the game time, to be allocated to the GM. Under the same match conditions, he gets to use more time for his moves, instead of pondering. The GM's time is more expensive, so it's reasonable to make the most, out of it.
Of course such a match is very much HW dependent, but as long as the system specs are published, anyone interested can gauge what it means. The general public, who doesn't care about such details, would be more impressed by the TC, than anything else. Even today, the common chess players, has no idea just how strong engines are, even at blitz/bullet. They all learned years ago, that machines are better at long TC, it's about time they learned a new lesson.

I am not sure they do not know that engines are very strong at short time controls, in fact it's likely that most chess players had several games against blitz settings rather than long control. I tested the more reasonable "throttling" of Komodo handicap at long time control (throttled to 10 kN/s or 1993 level 486-50 MHz). As a "human" contender I took Shredder 6 Paderborn, which on full modern core is pretty well calibrated to have a FIDE rating of about 2750 at tournament time control. I used long time control of 90 minutes + 15 seconds for both.



NPS are highlighted for both engines in the picture. The result after 12 games for the throttled Komodo: +6 -2 =4 against full Shredder, so Komodo on 1993 PC hardware is a bit above FIDE 2750. This roughly 100 speed (or time) handicap would be good enough to use against GM Movsesian, still with large chances to beat him at regular chess. If the games are with human at 45'+15'', then Komodo would need 27''+0.15'' on one modern core to be top GM level. But throttling to 1992-93 hardware seems better, because the feeling that the engine moves instantly is probably unsettling. Also, one can compare directly with 1992-93, when on the same hardware top engines were barely weak FM level. And one can separate the improvement during the last 22 years in FIDE ELO: about 500 software, about 400 hardware. All in all, from 2300 to 3200 FIDE ELO at tournament time control.

So based on that, if Komodo ran on my 24 core machine, which I'll guesstimate is about like a ten to one speedup, it would have to be given so little time that the amount of move overhead (default 30 ms) allocated would be the dominant factor in the result! I agree that throttling would be better, perhaps with just a two to one actual time limit to keep the total game time down to minimize tiredness. The influence of the opening book is also a major consideration, books now are vastly superior to books of 1993. Probably the most practical match we could hold from the normal starting position against a top GM would be Komodo on a cellphone (similar to throttling), playing Black, and with no or just three-move book. There is at least some commercial and spectator interest in a cellphone match, much more than in a "throttling" match.

[Engin]

such odd handicap games are total irrelevant for chess, i will even say its nonsense chess.

its making more sense to me if one grandmaster is want to play against a similar strength engine like about 2700 elo instead of 3300 elo.