Fritz 3200 strength does not crush me that quickly. Strange.

Sort:
HGMuller

They do not correct for the number of cores, because they typically test with a single core. When they run engines with multiple cores (2 or 4), they consider it another engine/version, and it is separately listed.

I don't know how the SSDF does their stuff.

I always found doubling the speed (or time) gives you closer to 70 Elo. Remember that doubling the number of cores is not the same as doubling the speed.

How increasing the computer rating works out against humans is not clear. Speed increases makes the computers tactically better, and comp-comp games are all about tactics. But as soon as it gets tactical, a human will get fried by a computer anyway. Humans beat computers by avoiding tactics, and exploiting their strategic stupidity. It is not at all sure the tactically better computer will be strategically less stupid.

intrepidattack

I think the problem is simple.

Your computer is much weaker than any server you're playing against online.

I'm willing to bet that my Deep Fritz 12 Engine running on a dual-core using 4GB or RAM would crush you quickly regardless of whatever rating it says it is (3000+).

The software is limited by your hardware.

Elroch

You are incorrect to suggest a strong program on an average PC will be weak, intrepidattack. As I have pointed out a couple of times, strength increases rather slowly with increased processing speed. Hence a strong program on a slow PC is still rather strong.

Chesserroo2

I had Fritz 8 play against my favorite online program. Fritz 8 checkmated it in the middle game. Fritz 8 crtiqued its move turn by turn, suggesting alternatives for 1/3 of them. At one point chess coach came on and said it thought a move was a bad idea and should be taken back. I told it to continue playing.

No matter what level I set the online program at, it always tends to make its moves in half a second, and instantly plays book moves. Fritz makes its moves in 15 seconds, including the very first move, indicating it is not using a book. Still, but how fast the online program responds, it seems that those moves are book.

Now that I think about it, 3200 does beat me faster than the online program does. It's opening moves just are not as in my face. It still checkmates me in the middle game, whereas the online program wins a few pieces off me and mates me in the endgame.

Elroch
HGMuller wrote:

They do not correct for the number of cores, because they typically test with a single core. When they run engines with multiple cores (2 or 4), they consider it another engine/version, and it is separately listed.

<snip>

I always found doubling the speed (or time) gives you closer to 70 Elo. Remember that doubling the number of cores is not the same as doubling the speed.

How increasing the computer rating works out against humans is not clear. Speed increases makes the computers tactically better, and comp-comp games are all about tactics. But as soon as it gets tactical, a human will get fried by a computer anyway. Humans beat computers by avoiding tactics, and exploiting their strategic stupidity. It is not at all sure the tactically better computer will be strategically less stupid.


Yes, I was referring to the listings of multi-core engines.

The rating increase for doubling speed seems poorly documented. My figure may have been nearer the low end of the estimates out there. I suspect the increase in rating goes down slowly as the strength gets higher (but this needs empirical checking).

Yes, doubling cores is a little less than doubling CPU speed, as some power gets wasted.

While relying on computer's "strategic stupidity" is a good idea, it has not been enough to world champions to prevail against computers. It is very difficult to unravel deep complex tactics from strategic judgement - computers manage to do things with calculation that humans can do with judgement (eg deep sacrifices). All strategic games are resolved eventually by some sort of tactical coup, and seeing these as early as possible is obviously very important.