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

I've played against a weaker chess engine and got ripped apart faster by more agressive play. That engine was educational.

I bought Fritz 8, and played against it in the unleased mode. I was amazed at how it was not putting pressure on my position or winning any material for quite a while. Is 3200 mode just very cautious/conservative? It did catch my blunders when I made them. After winning a knight, it decided to trade several pieces off the board and lock my pieces out of the queen side, all in a matter of 5 moves, and then it set its a pawn down to get queened and mate my king. I guess from that point on it was fast enough. But leading up that point I was quit unaware of its playing strength.

What is your experience with that playing strength?

I think another program I play against just assumes I won't see 5 moves ahead, and thus takes risks in order to beat me faster. I'd like to have that program play against the 3200 one.

Fritz 8 is not even close to 3200.

While I highly doubt Fritz 8 could play at 3200 strength (especially on a home computer), chess engines do have their quirks, maybe thats one of them in this case.

Edit: Too slow lol

Like you said, maybe the weaker engine took risks that just happened to pay off. The stronger engine just plays the board and gets small advantage after small advantage until the game is over.

Well, what is stopping you? Fritz GUI can play engines against each other, and accepts any engine using a standard protocol, right?

Yes, Fritz only beat Kramnik 4-2 on a machine with quite similar power to a typical modern computer, so it's probably not much over 3000.

You just seem to nail this thing every time.

You are my hero.

The trick is to just put it everywhere.

Quoted for truth.

If you want to play 3200, you need to play Houdini/DeepRybka4 on a powerful PC.

No thanks there are plenty of human opponents who are strong enough to beat me and whats more some of them ( not all) talk.

You have to remember that a good position will have no refutation to it, no matter the opponent. But it will be extremely punishing for the slightest blunder, wherever it comes. I have noticed times where I got pretty far against Houdini and Fritz 10; other times I would lose quickly. If you play for a really closed game, you probably won't get destroyed until things open up

I thought the 2400 figure sounded a suspiciously low guess, based on:

1. the fact that modern PCs compare rather well with even a high spec PC from 2006 (when Kramnik played)
2. Fritz 10 is undoubtedly stronger than Fritz 8, but not a huge amount
3. Doubling processor speed (or number of cores) increases Elo rating only by about 40 points.

Looking back at old computer rating lists gives Fritz 8's single core rating as about 2800 and it drew a match against Kramnik in 2002 (both on the slow single core PCs of the time).

Hence, even with a typical modern processor whose cores might be 10 times faster, the rating will probably be between 2900 and 3000 (the other side of 3000 from my earlier guess).

You are assuming Fritz 8 will be using more then one core, most likely it wont unless it's been optimise to. A program built to run on one core (normally older programs were) generally doesn't use more then one core.

The computer I'm using is a 2.2 GHz Pentium 4 from around 2004 or so, and has 512 MB of ram, running on Windows XP.

I don't know how to get the engine for the other program, which is an online free program that is used as an applet each time I go to the site. I'd download it if I could.

While it is true that I must blunder first in order to be crushed, I've noticed I blunder much sooner against the online program than I do against Fritz. With Fritz, I tend to have an even position by move 15. With the online program, I'm hanging on for life at move 8.

I also noticed I blunder more against Fritz's 1680 strength than against its supposed 3200 strength. All I can figure is the weaker strengths pick more !? moves whereas the 3200 strength avoids any risky moves.

On the contrary, I was assuming Fritz 8 would only be using one core. The cores were only considered for comparison with stronger multicore programs such as Deep Fritz 10.

In todays multicore computers, a single core's speed (from the multicore) is about as much or less then that of a single core several years ago. It's not unheard of to have single core programs from the past to actually run worse on modern computers. So I would guess that Fritz 8 is either as good or worse on modern computers then it was on computers it was designed for.

Edit: Though it doesn't seem to be the case here.

You are correct for the last few years, but 2002 is a long time ago, when the fastest single core CPU achieved about 3500 MIPS. Current top processors achieve around 25000 MIPS per core, with the speed advantage over my 4 year old dual core Athlon X2 machine being mostly due to rather higher clock speeds (and more cores). But that is fast compared to high-end 2002 machines when clock speeds were not much over 1Ghz.

[This still supports the point that Fritz 8 is nowhere near 3200 on a typical modern PC. Probably nearer 2800 than my first guess of 3000]

Note that all computer rating lists I know are made in a hardware independent way. That is, engines that had a rating of 2800 when they were tested on a 800MHz Hz single-core Pentium III in 2004 are still rated 2800 now that low-end computers typically have two cores, each twice as powerful in terms of instructions per cycle, running at 2.66 GHz. Rating testers simply adapt the time per move to the capabilities of the hardware.

The computer rating lists are just for comparing the strength of programs relative to each other. Not to give an absolute scale that could be compared to human rating lists.

While you are correct about the nature of the ratings lists, in cases where there are matches against humans, there is evidence that the ratings are quite similar. Bear in mind the fact that ratings only change quite slowly with speed (about 40 points for a doubling) moderates the importance of speed and core count.

CCRL does indeed use a standard CPU AMD Athlon X2 4500+ (previous generation to mine, so a few years old now. Maybe chosen in 2005 when CCRL started). But I believe that it does not correct for the number of cores, because the version of Crafty used for benchmarking only runs on only one core as far as I know.

While SSDF uses a standard CPU, many of the engines are stated to have been run on other CPUs, some only run on special platforms, and there is no indication that there is a correction for this in the time control.