Performance ratings of Komodo in GM handicap matches

[Lyudmil Tsvetkov]

...but I must say I'm a bit surprised that Stockfish improvement seems to have actually accelerated this year. I wonder why?

Very simple.

Look at the number of tests pushed=1.5/2 times more on average than for the year before.
- SF framework is practically never empty
- more testers in there
- much more cpu; as of late I saw over 100 machines/700 cores!! trudging under SF software patches; last year 500 cores was the record and framework was often empty or SF had problems with part of its Windows machines
- Bill (Henry) has pushed all possible(and some impossible) variations of search routines;

If you wonder why Houdini is doing very well at TCEC, actually better than Komodo at this particular time, the naswer is again very simple: Robert Houdart has been working(though not publishing) on his engine for 2 or 3 years, so by now he already has an engine at roughly Komodo/SF strength, but this is the farthest he has gone and will go in the near future.

So, a 3-way tie at the top currently, anyone could win TCEC, it will be a matter of luck mostly.

[Nordlandia]

I think in the variation "Transcendental Chess" this fen is one of the possible starting positions.

[d]qnbrrbnk/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w - - 0 1

I may be wrong though

https://en.wikipedia.org/wiki/Transcendental_chess

But i agree it's better to stick close in touch with standard chess.

[Lyudmil Tsvetkov]

I think in the variation "Transcendental Chess" this fen is one of the possible starting positions.

[d]qnbrrbnk/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w - - 0 1

I may be wrong though

https://en.wikipedia.org/wiki/Transcendental_chess

But i agree it's better to stick close in touch with standard chess.

It is better to stay in touch with common sense, as this has nothing to do with the thread topic.

[JJJ]

...but I must say I'm a bit surprised that Stockfish improvement seems to have actually accelerated this year. I wonder why?

Very simple.

Look at the number of tests pushed=1.5/2 times more on average than for the year before.
- SF framework is practically never empty
- more testers in there
- much more cpu; as of late I saw over 100 machines/700 cores!! trudging under SF software patches; last year 500 cores was the record and framework was often empty or SF had problems with part of its Windows machines
- Bill (Henry) has pushed all possible(and some impossible) variations of search routines;

If you wonder why Houdini is doing very well at TCEC, actually better than Komodo at this particular time, the naswer is again very simple: Robert Houdart has been working(though not publishing) on his engine for 2 or 3 years, so by now he already has an engine at roughly Komodo/SF strength, but this is the farthest he has gone and will go in the near future.

So, a 3-way tie at the top currently, anyone could win TCEC, it will be a matter of luck mostly.

Robert said he started working on houdini at the end of last year only. He also said he was back because he was thinking about surprise us all. So maybe he is right and his Houdini is strong enough to reach superfinal and maybe beat Stockfish if he improves Houdini enough during the stage 3.

But it is also too soon to be sure about the true strenght of this Houdini dev to me.

[lkaufman]

I think in the variation "Transcendental Chess" this fen is one of the possible starting positions.

[d]qnbrrbnk/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w - - 0 1

I may be wrong though

https://en.wikipedia.org/wiki/Transcendental_chess

But i agree it's better to stick close in touch with standard chess.

Chess960 is the most popular variant of chess with tournaments, ratings, engines that play it, online play etc. But who plays "transcendental chess"?

[Lyudmil Tsvetkov]

...but I must say I'm a bit surprised that Stockfish improvement seems to have actually accelerated this year. I wonder why?

Very simple.

Look at the number of tests pushed=1.5/2 times more on average than for the year before.
- SF framework is practically never empty
- more testers in there
- much more cpu; as of late I saw over 100 machines/700 cores!! trudging under SF software patches; last year 500 cores was the record and framework was often empty or SF had problems with part of its Windows machines
- Bill (Henry) has pushed all possible(and some impossible) variations of search routines;

If you wonder why Houdini is doing very well at TCEC, actually better than Komodo at this particular time, the naswer is again very simple: Robert Houdart has been working(though not publishing) on his engine for 2 or 3 years, so by now he already has an engine at roughly Komodo/SF strength, but this is the farthest he has gone and will go in the near future.

So, a 3-way tie at the top currently, anyone could win TCEC, it will be a matter of luck mostly.

Robert said he started working on houdini at the end of last year only. He also said he was back because he was thinking about surprise us all. So maybe he is right and his Houdini is strong enough to reach superfinal and maybe beat Stockfish if he improves Houdini enough during the stage 3.

But it is also too soon to be sure about the true strenght of this Houdini dev to me.

If you go by what Robert said...

You do not improve an engine by 130 elo in 6 months' time.

[Laskos]

7. Material seems to help humans more than moves, when comparing different handicaps with similar engine evals.

We want to measure the relative strength of the players at something resembling the game in question.

These two reasonable statements lead me to consider this hilarious handicap against a 2500-2600 GM:

[d]rnbq1bnr/pppppppp/8/4k3/8/8/PPPPPPPP/RNBQKBNR w KQ - 0 1

It was tried against a 2200 FM, and human lost. The total score of "hilarious positional handicaps vs FM" was 3.5/4 for Komodo, so the next step would be to try them against a GM. Humans do indeed feel more comfortable with more material rather than with huge positional advantage against engines. This particular position results in 3-4 tempi and a pawn down for Komodo, and would be a match for a 2500-2600 GM. If GM doesn't go for pawn, he will have to build up a huge positional advantage, which, it seems, more a stronger player thing, and human against Komodo can hardly do it.

[lkaufman]

7. Material seems to help humans more than moves, when comparing different handicaps with similar engine evals.

We want to measure the relative strength of the players at something resembling the game in question.

These two reasonable statements lead me to consider this hilarious handicap against a 2500-2600 GM:

[d]rnbq1bnr/pppppppp/8/4k3/8/8/PPPPPPPP/RNBQKBNR w KQ - 0 1

It was tried against a 2200 FM, and human lost. The total score of "hilarious positional handicaps vs FM" was 3.5/4 for Komodo, so the next step would be to try them against a GM. Humans do indeed feel more comfortable with more material rather than with huge positional advantage against engines. This particular position results in 3-4 tempi and a pawn down for Komodo, and would be a match for a 2500-2600 GM. If GM doesn't go for pawn, he will have to build up a huge positional advantage, which, it seems, more a stronger player thing, and human against Komodo can hardly do it.

It was also tried in a couple blitz games vs. GMs (Lenderman and maybe also Mikhalevsky) with losses for the GMs. FM Gilden did draw the similar position with K on d5, so Komodo performed about 2400 USCF or about 2300 FIDE in those two games, and he did so without even using a computer to prepare. The main problem with these positions is that the best lines of play are pretty much forced so too easy to prepare using a computer, and as you say result in positions similar to pawn and three move handicap. There is also the objection that the positions could not arise by legal play; I think that all the positions we have used in these matches other than the two in question could be reached by a legal sequence.

[Dirt]

... There is also the objection that the positions could not arise by legal play; I think that all the positions we have used in these matches other than the two in question could be reached by a legal sequence.

Yes, that does offend my esthetic sense. You could give P-K3 to both sides, though.

[Laskos]

... There is also the objection that the positions could not arise by legal play; I think that all the positions we have used in these matches other than the two in question could be reached by a legal sequence.

Yes, that does offend my esthetic sense. You could give P-K3 to both sides, though.

That would be more like handicap to give to 2700-2800 GM. It basically leads to pawn f7 grab and Queen exchange. A large positional advantage (3-4 tempi + castling) from hilarious legal starting position to a 2700-2800 GM is just a bit different:

[d]rnbq1bnr/ppppp1pp/5p2/4k3/8/4P3/PPPP1PPP/RNBQKBNR w KQ -