Author: Matthew Hull
Date: 10:49:07 02/19/04
Go up one level in this thread
On February 19, 2004 at 13:29:56, Dann Corbit wrote: >On February 18, 2004 at 21:32:03, Bob Durrett wrote: >[snip] >>Either you, Dann, or Bob Hyatt said something about whether or not a particular >>block of code might be well suited to "hardware-izing." The idea is that >>certain blocks of code could be easily "hardware-ized" and other blocks of code >>would be hard to "hardware-ize." > >That's the standard approach. No hardware solution ever has been pure hardware. > You also have to write software infrastructure. > >>Perhaps the better approach would be to break up the chess engine into >>functions, and measure percentage of microprocessor time used for each function. > >They call that "profiling" and every chess author does it. > >> There would, surely, be many ways to decompose the overall "chess engine >>function" into a set of sub-functions, and perhaps some innovativeness would >>help to make sure most of the small functions were well suited to >>"hardware-izing." Those that were not would require that the engine designer >>select alternatives to get around the difficulties. All this might take some >>effort. > >Suitablity for hardware will depend largely on the task. If it is easy to write >a circuit to perform a task, then it is more suitable to create a hardware >solution. > >>Your idea of working mainly on eliminating "bottlenecks" seems a good idea too. > >It's the only logical way to proceed. BTW, they have been doing that since the >1950's. > >>I still see the potential for enormous gains in engine strength, at least >>hypothetically. The proof would be in "the pudding," of course. Only when the >>hardware were built and tested would the performance gain be measured and >>understood. > >I think that Deep Blue, Hydra and other systems have already proven the concept. > We don't need to guess to know if it works or not. > >>Hydra may, indeed, be just a "flash in the pan," but you must admit that Hydra >>winning Paderborne should have raised a few eyebrows. Someone would have to be >>really insensitive to not be at least a bit pulsed. > >I think it is an interesting development. It's not a revolution by any stretch >of the imagination. It's an old technique, revived once again. Tying a pretty >blue marketing bow on top isn't new either. I think it's much more interesting have a software design where you can just throw CPUs at it, much as NUMA Crafty is today. There is much more potential for continuously, ever-increasing, deep searching growth there. I believe Bob once threatened Vincent (Vincent was bloviating on 512 processors or something) that he might just do that some day (a Crafty cluster, if you will), just to show him. Then you could throw a thousand or more CPUs at it and see if -- in the famous words of Hsu -- "That ought to do it". :)
This page took 0.01 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.