WCCC June 27th in Leiden

[diep]

Just got back from the pub so sorry for the late blitz final results

1. Jonny
2. Shredder
3. Komodo
4. Hiarcs

Wow, so he's using too many cores now!

Congratulations to Jonny!

And this where he plays every game some solid openingsline - Jonny really had to fight for every point it makes. No openingsadvantage whatsoever...

Komodo is outsearching jonny so i heard from Erdo. Is that correct?

I heard a search depth of 28 plies in one of the middlegames for Jonny.
Can someone give some comparisionnumbers there?

If he still searches similar to a few years ago i know Jonny isn't searching as thin as others do.

Maybe Jonny has some speedup guesses at 2400 cores?
Really seems a lot to me 2400.

I had big problems to get 500 cpu's going back in 2003...

The solution that solved it overnight basicaly (after round 7, so it only scaled ok rounds 8..11) meant that i partitioned it in 4 groups of 125 processors (effectively 115).

That reduced the bottleneck a lot. So basically without partitioning the load balancing went ok up until 120 cpu's roughly. Above that partitioning was needed back then.

2400 cores seems really a lot to me there

Assuming this is first time Jonny uses that many - any speedup numbers from Jonny?

[diep]

En passant bug in Komodo?

1.d4 d5 2.e4 fxe4 3.Nc3 Nf6 4.Bg5 Nc6 5.d5 Ne5 6.Qd4 Nf7 7.h4 e5?!

To me looks like dubious move. You just take on e6 en passant and whites (presumably bad position according to theory from 1990s) looks suddenly ok for white.

After 7..c6 - black is obviously much better.

I was hoping you were right so we could gain some elo points by a simple bugfix. But no, there is no bug. Komodo evaluates the endgame after the queen trade as equal, whereas it shows a tiny White plus after 7...c6. You say "black is obviously much better" but my database stats favor White as does Komodo, and it's not at all obvious to me which side is better.

Database with games of my teammember who managed to blow this a lot with black by blundering

Allowing the En Passant for sure is a very bad move. c6 really only move.

For sure some sort of hashing bug that only shows itself if you have tons of cores that already were searching. They do not know much about massive parallel correct search in computerchess.

It's clearly not a bug because if you search the positions after the queen trade and after 7...c6, Komodo marginally prefers the 7...e5 line. I'm not saying you are wrong about 7...c6 being the best move, but if so it means that Komodo's eval is a bit off in this particular situation; that isn't a bug.

That sure would be the fundamental explanation behind it Larry. Yet even then - search should correct evaluation here though Larry. It's too simple to evaluate all this for engines later on in the search, if you search tad correctly that is.

Getting 35 plies of search depth and then doing some sort of supernullmove that reduces 10 plies or so of course is madness

c6 is winning for black and e5 is a crap move, just Raptor played suicide chess after e5. Nb5?? for example. Blundering its better position to a worse one.

Though i didn't investigate whether the move e6-e5 later on was as bad as it seems versus Rf8 there.

[diep]

So maybe the next Shredder will have a very big elo jump. I also see it played a very nice game against Komodo in this WCCC where Komodo was to me close to loose.

Don't take me wrong - but the level i see in the games i wouldn't call that "big progress" past few years for computerchess.

[Laskos]

2400 cores seems really a lot to me there

Assuming this is first time Jonny uses that many - any speedup numbers from Jonny?

What speed-up? It gets 1 billion NPS. Don't know effective speed-up. Mark says it ponders many moves simultaneously, but ponder alone cannot mean more than twice the thinking time. Pity this thing is not tested in 100 games against Komodo, to have a better idea what this is.

[diep]

Shredder is greatly improved since last year, and is running on a 32 core machine. It is showing a lot of depth and it is clear Stefan has made a lot of progress.

Jonny also is much stronger. For instance, it is currently in second place on the TCEC round 2. Johannes Zwanzger, Jonny's programmer, says its performance in WCCC and speedchess on the cluster depends a lot on how lucky it gets on its pondering choices. Jonny ponders on multiple opponents moves. So you can some games where it is amazing and others where it gets draws against probably weaker opponents. But the core program does look a lot stronger to me.

The Software Championship this weekend has everyone running on identical machines, whihc will take out the hardware components in WCCC. It should be very interesting. I am happy to see all this hard work by all of these authors.

You forget to add that Jonny wins all this despite playing very solid openings. It never is trying trick openingslines unlike Erdo.

Erdo really gambles and in 90% of the times he gambles right.

[diep]

2400 cores seems really a lot to me there

Assuming this is first time Jonny uses that many - any speedup numbers from Jonny?

What speed-up? It gets 1 billion NPS. Don't know effective speed-up. Mark says it ponders many moves simultaneously, but ponder alone cannot mean more than twice the thinking time. Pity this thing is not tested in 100 games against Komodo, to have a better idea what this is.

Good pondering can be very powerful. It reduces the time of your opponent in half not seldom at cricial moments.

So it reduces elo bigtime of your opponent in such case.

with 2400 cores i would say a clever choice from Jonny.

[diep]

Shredder is greatly improved since last year, and is running on a 32 core machine. It is showing a lot of depth and it is clear Stefan has made a lot of progress.

Jonny also is much stronger. For instance, it is currently in second place on the TCEC round 2. Johannes Zwanzger, Jonny's programmer, says its performance in WCCC and speedchess on the cluster depends a lot on how lucky it gets on its pondering choices. Jonny ponders on multiple opponents moves. So you can some games where it is amazing and others where it gets draws against probably weaker opponents. But the core program does look a lot stronger to me.

The Software Championship this weekend has everyone running on identical machines, whihc will take out the hardware components in WCCC. It should be very interesting. I am happy to see all this hard work by all of these authors.

Jonny is on second place because of luck, but of course, everyone sees it plays a lot stronger. Still , I feel Komodo is still the unchallenged number 1.

Openingsbook from Erdo is very strong yet i find it plays horrible chess that Komodo.
Just more of the same thing. Too little chessknowledge still.

[diep]

Shredder is greatly improved since last year, and is running on a 32 core machine. It is showing a lot of depth and it is clear Stefan has made a lot of progress.

Jonny also is much stronger. For instance, it is currently in second place on the TCEC round 2. Johannes Zwanzger, Jonny's programmer, says its performance in WCCC and speedchess on the cluster depends a lot on how lucky it gets on its pondering choices. Jonny ponders on multiple opponents moves. So you can some games where it is amazing and others where it gets draws against probably weaker opponents. But the core program does look a lot stronger to me.

The Software Championship this weekend has everyone running on identical machines, whihc will take out the hardware components in WCCC. It should be very interesting. I am happy to see all this hard work by all of these authors.

Good to hear your report Mark.

If it's true what you say - then obviously the problem this world champs is the openingsbook of shredder. it works ok against what they practiced against yet the openingschoices made seem not so clever for example against Jonny.

Playing slav with an engine like shredder is suicide of course against faster hardware.

[Laskos]

2400 cores seems really a lot to me there

Assuming this is first time Jonny uses that many - any speedup numbers from Jonny?

What speed-up? It gets 1 billion NPS. Don't know effective speed-up. Mark says it ponders many moves simultaneously, but ponder alone cannot mean more than twice the thinking time. Pity this thing is not tested in 100 games against Komodo, to have a better idea what this is.

Good pondering can be very powerful. It reduces the time of your opponent in half not seldom at cricial moments.

So it reduces elo bigtime of your opponent in such case.

with 2400 cores i would say a clever choice from Jonny.

Ponder is at most a factor 2 in thinking time on average (either decrease the opponent's time or increase yours), and at this level would mean some 30-40 ELO points.

[Werewolf]

Assuming this is first time Jonny uses that many - any speedup numbers from Jonny?

Jonny had a similar number of cores in the last WCC IIRC