Which is the weaker engine that can find the draw in game #2

[kgburcham]

WHAT DID YOU FOUND OUT?

very complicated so far---still working on it.

[pichy]

WHAT DID YOU FOUND OUT?

very complicated so far---still working on it.

Thanks, but Deep Blue made those moves in an average of 1 to 2 minutes. I thought that today's engines using normal P.C. were much stronger than Deep Blue.

PS: Can you imagine if Houdini or Critter 1.6a codes could be programmed into Deep Blue

Houdini running on a $1,000 desktop computer has an estimated ELO of around 3200, so it should smoke any human competitor. I'm sure GM's can get a win every now and then but I don't think a human can beat Houdini in a match of games. Kind of sad, really.

When Kasparov lost the rematch against Deep Blue in 1997, the computer chess landscape was much different. Look at the specs it took to beat Kasparov, 3.5-2.5. (from Wikipedia of course):

It was a massively parallel, RS/6000 SP Thin P2SC-based system with 30-nodes, with each node containing a 120 MHz P2SC microprocessor for a total of 30, enhanced with 480 special purpose VLSI chess chips. The effective speed of specialized chips can not be deduced from their quantity or frequency, so the above information does not help much to compare to a regular desktop pc. The only thing of use would be how many positions are reviewed per sec. (I think that number still supercedes what today's desktops achieve).

On top of that, chess engines look at x positions and evaluate them. The rules for this evaluation differ from program to program. So even if we could say Deep Blue II was faster than an i7 cpu, the newer engines might still yield surprising results.

This is pretty much why engine ratings are computed by actually making them play each other. Even with comparable hardware, you can't predict what effect the evaluation has. If we knew what evaluation was superior in advance, we'd just write that chess program and be done with it.

[rbarreira]

The only thing of use would be how many positions are reviewed per sec. (I think that number still supercedes what today's desktops achieve).

Deep Blue did about 200 million nps on average, but its parallel efficiency was around 10% according to the authors, which means an effective 20 million nps if it was single-threaded. The reason for such low parallel efficiency is the high number of parallel units that Deep Blue employed, and the fact that they didn't share transposition tables with each other.

So today's quad-core CPUs are almost as efficient as Deep Blue was, as they can get nearly 20 million with a very high parallel efficiency.

But as you pointed out there's still the question of how fancy its evaluation was...