It would be comical to see deeper blue 1997 in TCEC!

[S.Taylor]

The greatest version of the super monster that crushed Kasparov in chess and spirit, 1997.

If it was now, in stage 3 of TCEC, it would lose every single game, sometimes in a few moves. Every game would 0 to DB.

[Laskos]

The greatest version of the super monster that crushed Kasparov in chess and spirit, 1997.

If it was now, in stage 3 of TCEC, it would lose every single game, sometimes in a few moves. Every game would 0 to DB.

Not quite. It would have achieved in 30 games in this stage say 3-4 draws and maybe a win or two, especially against Hannibal and Protector.

[S.Taylor]

The greatest version of the super monster that crushed Kasparov in chess and spirit, 1997.

If it was now, in stage 3 of TCEC, it would lose every single game, sometimes in a few moves. Every game would 0 to DB.

Not quite. It would have achieved in 30 games in this stage say 3-4 draws and maybe a win or two, especially against Hannibal and Protector.

Those are only fractionally weaker than SF and Komodo.
Probably a lot stronger than Rybka 4.1.

(correct me if I'm wrong)

[Laskos]

The greatest version of the super monster that crushed Kasparov in chess and spirit, 1997.

If it was now, in stage 3 of TCEC, it would lose every single game, sometimes in a few moves. Every game would 0 to DB.

Not quite. It would have achieved in 30 games in this stage say 3-4 draws and maybe a win or two, especially against Hannibal and Protector.

Those are only fractionally weaker than SF and Komodo.
Probably a lot stronger than Rybka 4.1.

(correct me if I'm wrong)

400 ELO points can be a misleading gap, it still means 10% score of the weak machine. Hannibal and Protector seem the level of Rybka 3, if their updates are not very important.

[nimh]

Back in 2009 I analyzed moves of both Deep Blue versions 1996 and 1997. It turned out that their combined strength was around a tad higher than 2800 in terms of FIDE 2007. See page 8. http://www.chessanalysis.ee/summary450.pdf

Now, according to my research a few years later FIDE 2800 corresponds to ca 2300 on CCRL 40/40 list, as you can see on page 16. http://www.chessanalysis.ee/Quality%20o ... suring.pdf

TCEC uses 3 times slower time controls and ca an 11 times faster machinery - 33 times increase in search capability which according to the Houdart's formula (25 * log x) / log 1.3 should be equal to ca 330 ELO increase. The avreage rating of the participants of the stage 3 of TCEC is 3228.

The rating difference is above 1200 points, more than 99.9% expectancy for the stronger side. So, I'd conclude that even the 1997 version of Deep Blue would be lucky to score one draw.

[Laskos]

Back in 2009 I analyzed moves of both Deep Blue versions 1996 and 1997. It turned out that their combined strength was around a tad higher than 2800 in terms of FIDE 2007. See page 8. http://www.chessanalysis.ee/summary450.pdf

Now, according to my research a few years later FIDE 2800 corresponds to ca 2300 on CCRL 40/40 list, as you can see on page 16. http://www.chessanalysis.ee/Quality%20o ... suring.pdf

TCEC uses 3 times slower time controls and ca an 11 times faster machinery - 33 times increase in search capability which according to the Houdart's formula (25 * log x) / log 1.3 should be equal to ca 330 ELO increase. The avreage rating of the participants of the stage 3 of TCEC is 3228.

The rating difference is above 1200 points, more than 99.9% expectancy for the stronger side. So, I'd conclude that even the 1997 version of Deep Blue would be lucky to score one draw.

I find several flaws here. Fritz 8, 8+, Junior 8 drew Kasparov and Kramnik in 12 games on 2-8 cores. These engines are about 2700 on CCRL 40/40 on one core. Then what's this "Houdart formula"? It means simply 66 ELO points per doubling at arbitrary time control, which is silly. Taking into account diminishing returns, 33 means roughly 5 doublings, and they will be something like 60, 50, 40, 35, 30 ELO points from CCRL 40/40 to TCEC. A total of 215 ELO points or so from a 33x factor. All in all, significantly lower than 1200 ELO points difference between DB and Stage 3 TCEC. Still maybe 500-600 points.

[nimh]

No wonder humans performed so well - after all CCRL 2700 corresponds to ca 2880 FIDE - a piece of cake to overcome using anti-computer strategy. Remember that my comparison uses assumption that humans do not use such strategy, as it is impossible to render into a certain amount of elo points. Human vs engien games are irrelevant in this instance.

Could you explain why 600 elo? Where does that figure come from?

Even if Houdart formula is wrong in your opinion, the gap still is some 1000 points according to my data.

[Laskos]

No wonder humans performed so well - after all CCRL 2700 corresponds to ca 2880 FIDE - a piece of cake to overcome using anti-computer strategy. Remember that my comparison uses assumption that humans do not use such strategy, as it is impossible to render into a certain amount of elo points. Human vs engien games are irrelevant in this instance.

Could you explain why 600 elo? Where does that figure come from?

Even if Houdart formula is wrong in your opinion, the gap still is some 1000 points according to my data.

There are other smaller flaws in your argument. The hardware is 11 times faster than 4 CCRL cores in NPS, not effective speedup, which is roughly 5-6, so one doubling is out. Fritz 8+ is CCRL 2700 one core, but 2-8 cores is CCRL 2800. Assuming that DB 1997 (which actually won against Kasparov) is as strong as those engines which drew Kasparov and Kramnik, then DB 1997 is also CCRL 2800. TCEC Stage 3 is less than 200 ELO stronger than the average of 3228 on CCRL, so 3400 CCRL. Difference is 600 ELO points (CCRL 40/40). And 500 ELO points for Hannibal or Protector.

[M ANSARI]

I don't think that DB would be able to save a single game against any of the top engines. If you look at the play of DB games, it is obvious that it has an extremely weak evaluation in comparison with engines today. It would get picked apart almost immediately and no amount of hardware can save a game against the latest engines once the position is "lost". It just goes from bad to worse. My guess is that DB would not have a chance even if Komodo or SF was running on the latest Iphone or Ipad.

[Gurcan Uckardes]

I share mostly Kai's comments. And to enlarge the speculative discussion, let me ask at what ELO level Kasparov played vs DB to lose that badly. He was 2800+ at the time. DP had beaten him. Do you think it was worth 2800?