Deep Fritz 13 vs. Houdini - It's On!

These guys use ALL their time.

Lots to learn here, for us mere mortals.

I think it would be beneficial to add some more RAM to your system, or get linux to speed up your system. Or both. ;)

Not sure if you were commenting on my system. I have 16 GB of DDR3 Kingston RAM in the system currently. I have the computer using the Fritz maximum 2.6 GB hash.

I have plenty of Linux experience under my belt, and I would very much like to try Chessbase in the Wine emulator, maybe in a virtual machine, to see how it performs.

Using deep positional analysis, using the cloud analysis feature of Deep Fritz 13. Using the same engine, with an Intel i7 (8 core) CPU at 14,000 nodes per second.

What are HASH TABLES?

From the Deep Fritz 13 user manual:

Hash tables are memory areas in which the program can store positions and evaluations while it is calculating the moves of a game. If the program encounters the same position again, it can simply take the evaluation from the hash tables, rather than analysing the position all over again.

Hash tables increase the playing strength of the program considerably. This is especially true of tactically strong engines like Fritz, Junior or Shreddder. Some run at well over 500,000 positions per second, and will fill the hash tables very quickly. After that, the search slows down. This is not the case in a slower, positionally oriented program, which processes less positions per second, and takes much longer to fill the hash tables.

For slower time controls and deep analysis the engines need large hash tables. Tournament games with an average of three minutes (180 seconds) per move would ideally require over 256 MB for the hash tables. On blitz levels 16 to 64 MB is enough.

Some engines work best with hash tables sizes that are powers of two. This means that 64 MB of hash tables is much more valuable than 63 MB. Hash table sizes are set in the "Load engine" menu.

Some technical notes

The Windows operating system takes unused memory and uses it for internal purposes (e.g., to cache file and speed up the hard disk). If you set very large hash tables, Windows has to free the memory the first time you activate the engine. To do so, it writes the contents to the hard disk, closes files, etc. The process can sometimes take a while, and during that time the hard disk will run constantly.

After the memory required for the hash tables is free the hard disk should show no more activity (except for short seeks in the openings). This applies to subsequent games as well.

If there is constant disk activity during the game, then the hash tables you have set are too large. Windows has installed part of the tables on your hard disk, which slows the program down drastically. You will notice that the search depth remains very low.

If there is no disk activity at all when you first start an engine, this generally means that you have been too timid with the hash tables size. You may want to increase the size to get optimum playing strength.

If you are playing an important tournament game with large hash tables, you should start the engine once to force Windows to release the required memory. Press Alt-F2 to start the engine, and Alt-F2 again to stop it when everything is quiet.

Okay, Deep Fritz really likes:

12. h3?

...so I have committed to it.

Very unexpected. Also interesting that it evaluates it at 0.00. Maybe it couldn't find better than a forced draw?

It would be very cool if somehow this lead to new theory on the Grunfeld.

Uh oh, I hope Fritz is right on this.

I went ahead and started evaluating after Bxf3, gxf3 which is probably a certainty. Starting at 13. gxf3 Critter was more or less on board (-0.03) until it reached ply 24, when it suddenly dropped to -0.24 after an expected 13... c5.

I know you're still a few plies ahead of this, even as you started at move 11., but I hope it turns around soon!

Yes, Critter sees c5 as well. The point is to chip away at the center and make the White king uneasy. With the bishop aiming at d4, and through it to b2, it looks pretty tasty and well worth a pawn.

The other interesting line Critter sees, though only 0.00 is:

13... Nd7!? 14. Nxd5 c5 15. dxc5 Nxc5 16. Bc2 Bxb2 -- That's just something I would never have considered OTB.

Critter is now looking at another very interesting try for Black:

13...c5 14. dxc5 d4!? 15. Ne4 b6! -- trying to blast open the whole queenside.

I think you can safely assume gxf3 and start crunching from that position. What's Houdini's predicted line look like now? Is c5 still imminent?

If this game lasts 100 moves I won't be able to take the tension.

When I find  an unexpected resource for my opponent it tends to be the next move, the one after or (if I am really on form) the one after that. Being able to calculate a dozen (or two dozen) moves ahead gives such an additional sense of significance to each move made.

Not to mention the waiting.

Fritz shows c5 as the next move for black.

...and having made 13. ...c5 it is very apparent that black has the momentum.

This is genuinely exciting.

"Houdini loves to sacrifice material for piece activity and square control."

Are any of the engines able to report on these kind of metrics separately, rather than simply the overall evaluation?

Hmm, I wonder why they don't. It would be interesting to see (rather than having to calculate/infer it oneself).

Well it looks like we've got 14..Nc6, which will of course test White as a defender. Personally, I was hoping for 14..d4, as it looked just a little bit more wild. Not better, but maybe more interesting. 

I still want to know what Fritz had up its sleeve with h3 though. It must have seen something. 

Well I for one don't think h3 was good, but I'd still like to think there was a little more to it than that. There must be some reason why it thought Nd7 was better than Nc6. 

At this point, I'm still predicting a draw, but Houdini is going to make Fritz work HARD for it. 

I'm putting Deep Fritz on infinite analysis for the night; new move mid-morning tomorrow.

I'd like to challenge the winner to a game against Critter, if you're up to it.