TalkChess.com

bob wrote:

CThinker wrote:

rebel777 wrote: Like to add another reason, the world championship is about the program that plays the best chess. And since limited hardware reduces the playing strength of a program it's unacceptable.
Ed

There is a much longer thread on the subject of the 8-core limit.

We already have a much more extensive knowledge of equal hardware performance of engines, from CCRL, CEGT, etc. Rybka is the champion, by a very wide margin over Naum. Naum then itself has a big lead over the rest.

If you think about it, the WCCC now is not about the best chess playing system, but has turned into, the luckiest engine that could beat Rybka by preventing Rybka to use the advancements that it has brought to computer chess (cluster).

Instead of celebrating Rybka's advancements, it is being curtailed. Sad, really.

Now I can't wait for Bob's version of Crafty for clusters. The same for Naum. I hope to see a Naum for clusters soon. I'm not looking forward to WCCC to encourage the advancement of computer chess.

I don't buy the "this hurts Rybka" idea, because the cluster rybka is a joke. And a poor joke at that. There have been some decent cluster-based programs. But Rybka is simply not one of them.

Where did that come from ??

Vas

Vasik Rajlich wrote:

bob wrote:I don't buy the "this hurts Rybka" idea, because the cluster rybka is a joke. And a poor joke at that. There have been some decent cluster-based programs. But Rybka is simply not one of them.

Where did that come from ??

Vas

There is something of an explanation here.

Dirt wrote:

Vasik Rajlich wrote:

bob wrote:I don't buy the "this hurts Rybka" idea, because the cluster rybka is a joke. And a poor joke at that. There have been some decent cluster-based programs. But Rybka is simply not one of them.

Where did that come from ??

Vas

There is something of an explanation here.

I read this post and I can say 2 things.

1)I think that it is impossible to know the algorithm rybka is using based on output from a single position.

It is possible that something similiar that is not exactly the same is used
when some illogical moves that lose the queen are analyzed but this is not all the story and the algorithm is based partly on "split only at the root" and partly on another idea.

2)I remember that Vas said 100 elo based on testing at fast time control and I suspect that at fast time control you get clearly more than 50 elo per doubling so practically 5 nodes do not give 4:1 speed improvement but clearly less than it(maybe 2.5:1).

Uri

Vasik Rajlich wrote:

bob wrote:

CThinker wrote:

rebel777 wrote: Like to add another reason, the world championship is about the program that plays the best chess. And since limited hardware reduces the playing strength of a program it's unacceptable.
Ed

There is a much longer thread on the subject of the 8-core limit.

We already have a much more extensive knowledge of equal hardware performance of engines, from CCRL, CEGT, etc. Rybka is the champion, by a very wide margin over Naum. Naum then itself has a big lead over the rest.

If you think about it, the WCCC now is not about the best chess playing system, but has turned into, the luckiest engine that could beat Rybka by preventing Rybka to use the advancements that it has brought to computer chess (cluster).

Instead of celebrating Rybka's advancements, it is being curtailed. Sad, really.

Now I can't wait for Bob's version of Crafty for clusters. The same for Naum. I hope to see a Naum for clusters soon. I'm not looking forward to WCCC to encourage the advancement of computer chess.

I don't buy the "this hurts Rybka" idea, because the cluster rybka is a joke. And a poor joke at that. There have been some decent cluster-based programs. But Rybka is simply not one of them.

Where did that come from ??

Vas

RIght above my response:

If you think about it, the WCCC now is not about the best chess playing system, but has turned into, the luckiest engine that could beat Rybka by preventing Rybka to use the advancements that it has brought to computer chess (cluster).

Uri Blass wrote:

Dirt wrote:

Vasik Rajlich wrote:

bob wrote:I don't buy the "this hurts Rybka" idea, because the cluster rybka is a joke. And a poor joke at that. There have been some decent cluster-based programs. But Rybka is simply not one of them.

Where did that come from ??

Vas

There is something of an explanation here.

I read this post and I can say 2 things.

1)I think that it is impossible to know the algorithm rybka is using based on output from a single position.

It is possible that something similiar that is not exactly the same is used
when some illogical moves that lose the queen are analyzed but this is not all the story and the algorithm is based partly on "split only at the root" and partly on another idea.

2)I remember that Vas said 100 elo based on testing at fast time control and I suspect that at fast time control you get clearly more than 50 elo per doubling so practically 5 nodes do not give 4:1 speed improvement but clearly less than it(maybe 2.5:1).

Uri

The problem is, I am not basing my opinion on a single position. I am basing it on a whole game I watched in the last ACCA online event. Where Rybka was kibitzing PVs that showed _exactly_ how the search was being split and where...

Uri Blass wrote:

Dirt wrote:

Vasik Rajlich wrote:

bob wrote:I don't buy the "this hurts Rybka" idea, because the cluster rybka is a joke. And a poor joke at that. There have been some decent cluster-based programs. But Rybka is simply not one of them.

Where did that come from ??

Vas

There is something of an explanation here.

I read this post and I can say 2 things.

1)I think that it is impossible to know the algorithm rybka is using based on output from a single position.

It is possible that something similiar that is not exactly the same is used
when some illogical moves that lose the queen are analyzed but this is not all the story and the algorithm is based partly on "split only at the root" and partly on another idea.

2)I remember that Vas said 100 elo based on testing at fast time control and I suspect that at fast time control you get clearly more than 50 elo per doubling so practically 5 nodes do not give 4:1 speed improvement but clearly less than it(maybe 2.5:1).

Uri

The effective speedup is probably somewhere between 2.5:1 and 3:1 for 5 nodes, which is what Lukas had when he tested all of this.

Now he's up to 9 nodes BTW

Vas

bob wrote:

Uri Blass wrote:

Dirt wrote:

Vasik Rajlich wrote:

bob wrote:I don't buy the "this hurts Rybka" idea, because the cluster rybka is a joke. And a poor joke at that. There have been some decent cluster-based programs. But Rybka is simply not one of them.

Where did that come from ??

Vas

There is something of an explanation here.

I read this post and I can say 2 things.

1)I think that it is impossible to know the algorithm rybka is using based on output from a single position.

It is possible that something similiar that is not exactly the same is used
when some illogical moves that lose the queen are analyzed but this is not all the story and the algorithm is based partly on "split only at the root" and partly on another idea.

2)I remember that Vas said 100 elo based on testing at fast time control and I suspect that at fast time control you get clearly more than 50 elo per doubling so practically 5 nodes do not give 4:1 speed improvement but clearly less than it(maybe 2.5:1).

Uri

The problem is, I am not basing my opinion on a single position. I am basing it on a whole game I watched in the last ACCA online event. Where Rybka was kibitzing PVs that showed _exactly_ how the search was being split and where...

I think that you cannot get conclusions based on pvs that rybka showed.

It is possible that rybka gave only part of the information in the pvs that you saw and the algorithm is based on combination of splitting in the root and something else.

Uri

Uri Blass wrote:

bob wrote:

Uri Blass wrote:

Dirt wrote:

Vasik Rajlich wrote:

bob wrote:I don't buy the "this hurts Rybka" idea, because the cluster rybka is a joke. And a poor joke at that. There have been some decent cluster-based programs. But Rybka is simply not one of them.

Where did that come from ??

Vas

There is something of an explanation here.

I read this post and I can say 2 things.

1)I think that it is impossible to know the algorithm rybka is using based on output from a single position.

It is possible that something similiar that is not exactly the same is used
when some illogical moves that lose the queen are analyzed but this is not all the story and the algorithm is based partly on "split only at the root" and partly on another idea.

2)I remember that Vas said 100 elo based on testing at fast time control and I suspect that at fast time control you get clearly more than 50 elo per doubling so practically 5 nodes do not give 4:1 speed improvement but clearly less than it(maybe 2.5:1).

Uri

The problem is, I am not basing my opinion on a single position. I am basing it on a whole game I watched in the last ACCA online event. Where Rybka was kibitzing PVs that showed _exactly_ how the search was being split and where...

I think that you cannot get conclusions based on pvs that rybka showed.

It is possible that rybka gave only part of the information in the pvs that you saw and the algorithm is based on combination of splitting in the root and something else.

Uri

It is possible Rybka had a GM playing the moves and supplying nonsensical kibitzes as well. But it is not very likely. This is going nowhere. I _did_ an algorithm like that in 1983. I put it together in under 2 weeks and won the 1983 WCCC championship with it. And I _know_ how it works and how the PVs look. There's no doubt in this case as to what is going on. There is no other possible explanation. Feel free to offer _any_ reason how one could get a PV for depth-23, then a PV for depth=19, then a PV for depth=21, then a PV for depth=18, were if you take the best moves, and collect them individually, you can see one cluster node searching a group of moves in the normal way, kibitzing a PV after each iteration, another node doing the same for a different group of nodes. Etc. Or offer any explanation of why a program would kibitz (intermingled with the kibitzes of other nodes at different depths) a steady depth progression, with a score of -9.0, because there is only one possible recapture and that move is not being searched on that node.

This is _not_ guesswork. It is a simple statement of what it was doing. And how...

As far as for "why this was done" it is because it is an easy way to get a cluster search to work. But not effective. Not even +20 Elo effective...

Vasik Rajlich wrote:

Uri Blass wrote:

Dirt wrote:

Vasik Rajlich wrote:

bob wrote:I don't buy the "this hurts Rybka" idea, because the cluster rybka is a joke. And a poor joke at that. There have been some decent cluster-based programs. But Rybka is simply not one of them.

Where did that come from ??

Vas

There is something of an explanation here.

I read this post and I can say 2 things.

1)I think that it is impossible to know the algorithm rybka is using based on output from a single position.

It is possible that something similiar that is not exactly the same is used
when some illogical moves that lose the queen are analyzed but this is not all the story and the algorithm is based partly on "split only at the root" and partly on another idea.

2)I remember that Vas said 100 elo based on testing at fast time control and I suspect that at fast time control you get clearly more than 50 elo per doubling so practically 5 nodes do not give 4:1 speed improvement but clearly less than it(maybe 2.5:1).

Uri

The effective speedup is probably somewhere between 2.5:1 and 3:1 for 5 nodes, which is what Lukas had when he tested all of this.

Now he's up to 9 nodes BTW

Vas

Can we stay in the real world? Splitting at the root can not produce a 2.5x speedup, when the best move at the root takes _way_ over 50% of the total search time. There is theory. There is practice. And there is nonsense. For the event I am talking about, this claim is "nonsense". You might get the uninformed to buy this stuff, but not someone that has been doing it for 30+ years now (my first parallel search played its first ACM event in 1978....)

bob wrote:

Vasik Rajlich wrote:

Uri Blass wrote:

Dirt wrote:

Vasik Rajlich wrote:

bob wrote:I don't buy the "this hurts Rybka" idea, because the cluster rybka is a joke. And a poor joke at that. There have been some decent cluster-based programs. But Rybka is simply not one of them.

Where did that come from ??

Vas

There is something of an explanation here.

I read this post and I can say 2 things.

1)I think that it is impossible to know the algorithm rybka is using based on output from a single position.

It is possible that something similiar that is not exactly the same is used
when some illogical moves that lose the queen are analyzed but this is not all the story and the algorithm is based partly on "split only at the root" and partly on another idea.

2)I remember that Vas said 100 elo based on testing at fast time control and I suspect that at fast time control you get clearly more than 50 elo per doubling so practically 5 nodes do not give 4:1 speed improvement but clearly less than it(maybe 2.5:1).

Uri

The effective speedup is probably somewhere between 2.5:1 and 3:1 for 5 nodes, which is what Lukas had when he tested all of this.

Now he's up to 9 nodes BTW

Vas

Can we stay in the real world? Splitting at the root can not produce a 2.5x speedup, when the best move at the root takes _way_ over 50% of the total search time. There is theory. There is practice. And there is nonsense. For the event I am talking about, this claim is "nonsense". You might get the uninformed to buy this stuff, but not someone that has been doing it for 30+ years now (my first parallel search played its first ACM event in 1978....)

This happens frequently that older experts argue that something wouldnt or couldnt work because it never worked or had been done. There are some other excuses. If it were true, progress would end in smoke.