I was recently reading a post on a chess engine blog which sent me to a page that gave hints on using a chess engine in server play. Much to my surprise it turned out to be a link to the CCLA server page but they do NOT allow engine use. I assume that it was there because the CCLA is affiliated with the ICCF which DOES allow engine use. In any case there was much of interest in the advice and I am giving only the highlights here. If you want to read the entire article you can do so on the CCLA Server page.
To begin with, hardware and software specs that are used by top level “Advanced Chess” players are a secret…they won’t tell anybody how they do it, but it is known that the more powerful the computer the better. If you just want to dabble in engine assisted chess and don’t care about your rating a cheap computer will be OK, but be prepared to lose a lot and/or get stuck at low level because your computer isn’t packing the gear to take you any further.
You are going to need a machine capable of analyzing to about 40 ply, a dual-quad core processor, a 64-bit version of Windows and at least 8, preferably 16, gigs of memory. After running all this stuff for 8 hours or so, you will have generated a lot of heat so you have to know the maximum operating temperature of your machine and better yet, have it alarmed. Houdini 4 Pro, for example, can use 32 threads (16 cores) and 256 gigs of hash, but to get a machine at that level will cost you a bundle of cash. You can use your laptop, but don’t expect much.
For starters you must first configure the number of cores, hash size, endgame tablebases support and select the proper use of opening books. For Infinite Analysis configure maximum hash. That means that if your PC has 2048 MB RAM, then your maximum hash size is 1024 MB. For matches between engines it depends on what time control they play. Note: if you are conducting a Shootout in Fritz at 5 minutes per game then 128MB is enough because reading a large hash file while playing blitz is counter-productive. Engines are not good with openings or opening-related positions and it’s not a good idea to rely on the evaluation of positions up to about the 15th move.
It is better to analyze on one core for two hours than two cores for one hour. Adding a second core doesn't mean you add another 100 per cent to your analysis power. It means you add about 75 per cent to your previous 100. And these percentages diminish with every core you add because of the fact that the cores work in parallel and can often try to work simultaneously on the analysis lines. So, for analysis it's better to have one powerful core than two weaker ones. The exception would be if you are using Aquarium IDeA; you need more cores even if they are weaker.
Theoretically, the infinite analysis is better when using one powerful core for two days (with maximum hash) than using two cores for one day only. How long you allow your engine to examine a position depends. There is a horizon that engines can't pass, because of the pruning. i.e. they tend not to change their evaluations even if they spot something many plies from the root.
Concerning those sites that run engine vs. engine tournaments and establish engine ratings: First, remember that these engines only play against each other and so the strongest engine will win most of the time and just keep piling up points. Further, these engines are tuned specifically to compete in these tournaments, usually at blitz speed, so an engine that tops the list may not necessarily be the best for correspondence chess.
In Advanced Chess you can never, ever get in a hurry. One important thing is to never second guess the engine. That may have been possible in the old days, but not anymore (we are speaking of high level chess here). When an engine suggests two or three moves that are all evaluated about equally, there are some things you can try to ferret out the best move, but when all else fails, only then can you use your brain to choose the best move. Very rarely though will you reach a truly even position. When this happens the best thing to do is exit the multiple variation mode and analyze the top equally evaluated moves one at a time in the Infinite Analysis mode. In fact, this is good advice for all moves. If a move’s evaluation drops when you switch to single line mode, go back and do some more analysis. Sometimes the top rated move will not change no matter how many plies deep you go. One study of four popular engines found this to be the case 85 percent of the time, but at the top level of Advanced Chess that other 15 percent will kill you because when those guys play, 30 ply is not state of the art.
I have a Toshiba Satellite L855 running Windows 8, 2.50 GHz processor with 4.0 GB of RAM and a 64-bit operating system. I use Fritz 12 GUI, Stockfish 5, Houdini 2 and Critter 1.6 engines. For most of my analysis I use the Infinite Analysis mode and let the engines run from 10-15 minutes and with an occasional position, 4 or 5 hours. I joined Lechenicher SchachServer in 2008 at my CCLA rating of circa 2060, but wasn’t aware that they allowed engines so only scored +0 -4 =2 in my first tournament and lost a gob of rating points. When I realized they allowed engines I was using Fritz 5, only upgrading a couple of years ago, which lead to somewhat better results. To date my record there is only a +11 with 57 percent of the games drawn and a rating that has not varied significantly for a couple of years.
Short version: unless you want to invest a lot of money and spend just about all your waking time fiddling with your games, don’t bother playing correspondence chess unless it’s on a site that does not allow engine use. Even then, if you get up around 2200 all bets are off. If you just enjoy piddling around with different openings and various engines then Advanced Chess can be enjoyable…just don’t expect to gain a lot rating points. But, it can be fun doing like I did in a recent tournament. Knowing most guys are using pretty much the same equipment I am and many are playing a lot of games simultaneously so don’t let their engines analyze for a long time, I prepared some analysis in the crappy old Urusov gambit (1.e4 e5 2.Bc4 Nf6 3.d4) and was rewarded with a win and a draw! Who knows? I may research the Urusov deeper and try it again. Or maybe even seek out some other obscure gambit.
No comments:
Post a Comment