Author: Robert Hyatt
Date: 08:46:20 01/30/99
Go up one level in this thread
On January 30, 1999 at 01:45:31, KarinsDad wrote: >On January 30, 1999 at 01:03:36, Robert Hyatt wrote: > >>On January 30, 1999 at 00:45:27, KarinsDad wrote: >> >>>On January 29, 1999 at 22:15:15, Robert Hyatt wrote: >>> >>>[snip] >>> >>>> >>>>>>2) If it would be possible, not one line of code would be optimized to use that >>>>>>kind of hardware so the performance would be a disaster compared to Deep B. >>>>> >>>>>Yes and if the Deep Blue hardware had no C compiler written for it, you would >>>>>have to either write one or re-write the code (similar to what Bionic did) into >>>>>a supported language. And, you would have to re-optimize the code. That does not >>>>>mean that the basic algorithms of the current PC software could not be used and >>>>>would not be better. >>>> >>>> >>>>there is _no_ C compiler for the DB hardware. the chips are vlsi circuits >>>>and not something that is 'programmable'... >>>> >>> >>>The following quote comes from the following IBM web page: >>> >>>http://www.chess.ibm.com/meet/html/d.3.2.html >>> >>>The software inside of Deep Blue is one all-inclusive program written in C, >>>running under the AIX operating system. Deep Blue utilizes the IBM SP Parallel >>>System called MPI. "It's a message-passing system," says Hoane. "So the search >>>is just all control logic. You're passing control messages back and forth that >>>say, well, what am I doing? Did you finish this? OK, here's your next job. That >>>kind of thing at the SP level." >>> >>> >>>I also seem to recall that in the first match, the programmers tweaked the >>>software between games. How could this be done in hardware? Nobody would go back >>>to a game with newly burned chips that did not have significant tests. But a >>>small tweak of the software, piece of cake. >>> >>>What am I missing here Robert? >>> >>>KarinsDad >>> >>>[snip] >> >> >> >>A whole lot of hardware. :) >> >>DB has two distinct parts. A traditional chess engine running on the SP >>machine (C) code... >> >>And the chess-specific hardware that this C code uses to manage the board, >>plus do the last 4-5 plies of search + captures and that stuff, + the static >>evaluation. >> >>That part of DB is pure silicon. And anyone using it would have to write code >>similar to the part of DB running on the SP (ie a parallel search) that uses >>the special-purpose hardware to play chess. >> >>_that_ is why a commercial program can't be 'ported'. 'ported' is the wrong >>word, perhaps "morphed" is better, because the commercial program would suddenly >>look just like the current DB, necessitated by the DB hardware protocol >>requirements... > >So for the Bionic discussion, would it be that Bionic was "morphed" onto the >Crafty SMP/search code? > that's kind of 'backward'... ie the bionic approach is to modify the 'tips' of the tree, but use the same tree search code, unchanged. For DB, you have to modify the search part because you can't change the way the 'tips' are handled by the hardware. so there is a similarity, but it is sort of 'inverted'.. >Is the concept similar? I assume that the static evaluation in Deep Blue was >programmable by the software. No... the 'weights' are variable.. but the chips are 'hard-wired' according to Hsu, which means, that _nothing_ can be changed except weights. You could turn a term off by making the weight 0 of course, or make it bigger by scaling up the weight... but the 'code' (if you call gates code which is a stretch) can't be changed at all without a hardware revision of the chip... Otherwise, changing the weighing factors would be >a bear. Hence, it seems to me that any software could decide what is important >within the Deep Blue hardware as far as evaluation is concerned. Granted, >deficiencies in the PC software evaluation functions (such as the ones that use >minimal evaluations in order to acquire speed) would seem to "morph" into weaker >programs than programs whose evaluation functions are more robust. Of course, >since the evaluation functions are often tweaked to match the search engine >capabilities (and vice versa) for a specific program, all PC software "morphs" >would require additional tweaks to take advantage of the speed. > >KarinsDad and that's the point. They would likely end up with a 'deep blue' because those guys (DB team) are very sharp... and have had years to learn how to use such speed to best advantage. Once a commercial programmer (or even myself as an amateur programmer) got such hardware, we'd most likely re-discover the same things and end up doing things the same way... ie we all eventually end up with alpha/beta. we all eventually end up doing endpoint eval as the other approaches have accuracy problems. etc..
This page took 0 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.