Crafty on Cray Blitz vs. Rybka3

[Sean Evans]

Hi, if Dr. Hyatt had played his Crafty program by Internet hook up on the Cray Blitz system running at his University, would it be strong enough to beat Rybka3 on the hardware it was using by long distance?

Thanks

Sean

[fern]

I have the feeling that Rybka could crash Crafty even if this las run in Hall 2000 and Rybka runs in an Atari 1029

Fern

[Uri Blass]

Crafty has no chance even if get 30:1 hardware advantage
Rybka is very very strong.

The elo gap between rybka and Crafty is clearly more than 400 elo.


Uri

[CRoberson]

Crafty has no chance even if get 30:1 hardware advantage
Rybka is very very strong.

The elo gap between rybka and Crafty is clearly more than 400 elo.


Uri

I beg to differ. If you look in the CCC tournaments section, you'll
see a match between Telepath and Revelation Rybka which Telepath
won. The Revelation board has Rybka 2.2 running at 500 MHz while
I had Telepath running in 64 bit mode on 1 cpu of a Q6600 running
at 3.2 GHz. I think speed did help.

The speed up for Telepath going from 500 MHz 32 bit to 3.2 GHz
and 64 bit is about 12 x. This shows that with a 12x speed advantage
Telepath could beat Rybka - albeit Rybka 2.2.

However, his question pertains to the 40 cpu cluster and Crafty on
a Cray. Why not Crafty on a big cluster - Bob has one bigger than
40 nodes at UAB and each node has 8 cpu's.

The issue then will be network saturation. That many nodes could
saturate a network quickly. However, the UAB cluster uses infiniband
for network communication.

[Uri Blass]

Crafty has no chance even if get 30:1 hardware advantage
Rybka is very very strong.

The elo gap between rybka and Crafty is clearly more than 400 elo.


Uri

I beg to differ. If you look in the CCC tournaments section, you'll
see a match between Telepath and Revelation Rybka which Telepath
won. The Revelation board has Rybka 2.2 running at 500 MHz while
I had Telepath running in 64 bit mode on 1 cpu of a Q6600 running
at 3.2 GHz. I think speed did help.

The speed up for Telepath going from 500 MHz 32 bit to 3.2 GHz
and 64 bit is about 12 x. This shows that with a 12x speed advantage
Telepath could beat Rybka - albeit Rybka 2.2.

However, his question pertains to the 40 cpu cluster and Crafty on
a Cray. Why not Crafty on a big cluster - Bob has one bigger than
40 nodes at UAB and each node has 8 cpu's.

The issue then will be network saturation. That many nodes could
saturate a network quickly. However, the UAB cluster uses infiniband
for network communication.

one game mean nothing and you clearly did not use Rybka3 and even not rybka2.3.2a.

I also suspect that you do not estimate correctly the hardware difference
and it is bigger than it and you cannot use mhz as estimate for speed
because there are other factors that are more important for speed.

I also do not know what is the strength of Telepath compared to Crafty on equal hardware.

I know that rybka2.3.2a 32 bits can score more than 50% against movei with 10:1 time handicap based on many ponder off games when I give movei 40/40 against 40/4 of rybka time control so probably Rybka3 can beat movei with 30:1 time handicap and movei is stronger than Crafty on equal hardware based on rating lists.

Uri

[BubbaTough]

one game mean nothing and you clearly did not use Rybka3 and even not rybka2.3.2a.

I also suspect that you do not estimate correctly the hardware difference

Well on ICC LearningLemming only gets the occasional draw against Rybka 3 on good hardware, but has a plus score against Rybka 3 on really bad hardware. So clearly it has some effect. How much of an effect is debatable of course, and there are probably diminishing returns. All that said, no I don't think a big Crafty cluster would be competitive with the Rybka cluster.

-Sam

[Uri Blass]

Crafty has no chance even if get 30:1 hardware advantage
Rybka is very very strong.

The elo gap between rybka and Crafty is clearly more than 400 elo.


Uri

I beg to differ. If you look in the CCC tournaments section, you'll
see a match between Telepath and Revelation Rybka which Telepath
won. The Revelation board has Rybka 2.2 running at 500 MHz while
I had Telepath running in 64 bit mode on 1 cpu of a Q6600 running
at 3.2 GHz. I think speed did help.

The speed up for Telepath going from 500 MHz 32 bit to 3.2 GHz
and 64 bit is about 12 x. This shows that with a 12x speed advantage
Telepath could beat Rybka - albeit Rybka 2.2.

However, his question pertains to the 40 cpu cluster and Crafty on
a Cray. Why not Crafty on a big cluster - Bob has one bigger than
40 nodes at UAB and each node has 8 cpu's.

The issue then will be network saturation. That many nodes could
saturate a network quickly. However, the UAB cluster uses infiniband
for network communication.

I looked at the ssdf list and see the following:

11 Rybka 1.2 256MB Athlon 1200 MHz 2902 25 -24 893 71% 2747
37 Revelation Rybka 2.2 2653 69 -62 126 68% 2519

I believe the speed of Revelation is clearly less than 500 mhz
It is probably something like 100 mhz.

50 Junior 7.0 256MB Athlon 1200 MHz 2689 22 -22 1008 51% 2680
68 Junior 7.0 128MB K6-2 450 MHz 2627 20 -20 1200 56% 2583

speed improvement of 500:1200 cannot give 250 elo improvement as showed by the Junior7 example and at ssdf time control with books we have only 50 elo per doubling the speed or something like that.

Uri

[Uri Blass]

one game mean nothing and you clearly did not use Rybka3 and even not rybka2.3.2a.

I also suspect that you do not estimate correctly the hardware difference

Well on ICC LearningLemming only gets the occasional draw against Rybka 3 on good hardware, but has a plus score against Rybka 3 on really bad hardware. So clearly it has some effect. How much of an effect is debatable of course, and there are probably diminishing returns. All that said, no I don't think a big Crafty cluster would be competitive with the Rybka cluster.

-Sam

I wonder if the plus score against rybka3 on bad hardware is based on blitz or on slower time control.
Based on my experience in handicapped comp-comp games slower time control usually helps the better player at eqaul time control.

Uri

[Jouni]

From CEGT and CCRL Crafty seems to be 500 points after R3. This means
10 speed doublings = 1024x speed-up to get even score!

Jouni

[BubbaTough]

I wonder if the plus score against rybka3 on bad hardware is based on blitz or on slower time control.
Based on my experience in handicapped comp-comp games slower time control usually helps the better player at eqaul time control.

Played many games at 15 1, not so much slower than 15 minutes. Regarding slower time controls, slower time controls is equivalent to faster hardware for both sides. So I guess you agree that increasing hardware has diminishing returns.

-Sam