Author: Robert Hyatt
Date: 06:05:35 02/01/02
Go up one level in this thread
On February 01, 2002 at 05:47:21, Ed Schröder wrote: >On February 01, 2002 at 00:24:07, Robert Hyatt wrote: > >>On January 31, 2002 at 15:19:38, Ed Schröder wrote: >> >>> >>> >>>Okay 9 plies it is, it does not matter. >>> >>>Here is a snip from the "IEEE MICRO" journal from 1999. It says 4 plies in the >>>hardware AS PART of the iteration, thus not ADD TO the iteration. The text is >>>below. >>> >>>Reading the June 2001 article I seriously doubt DB has shared memory for its >>>hash table although it's not entirely clear to me. If true that is about the >>>biggest search killer there you can imagine which makes a 16-18 ply (brute >>>force) search claim even more ridiculous. Text below. >> >> >>DB definitely did _not_ have shared memory. The SP2 has no shared memory, it >>is a very fast message-passing system. > > >Here you go.... > >Brute force, massive extensions, no shared memory. > >16-18 plies is impossible. > Based on your attempts to do a distributed program? Ever heard of a distributed memory hash table? Happens all the time. Works well. No reason why it would cause any problems for DB either. Sun Phoenix used such an approach, for example. Waycool also did on a hypercube. Perfectly doable. > > >>The 4 plies makes no sense to me in any context, as Deep Thought searched >>4-5 plies in hardware, while deep blue searched 5-7 according to Hsu. This >>depth is pretty-well fixed by the speed of the SP2. The deeper the hardware >>goes, the slower a search goes and the host software ends up waiting on the >>chess hardware. If the chess hardware searches too shallowly, say 4 plies, >>then the host software can't keep up. For a given position/depth, there is >>a very precise "hardware depth" that optimizes performance... >> >>All explained by Hsu several times of course... > >You snipped Hsu's own text that explains the opposite. It says 4 plies in the >hardware AS PART of the iteration, thus not ADD TO the iteration. So what... the log files don't show an iteration number. And hsu's exact statement about X(Y) was "the search does X plies in software, which has all the SE stuff and everything else, and then Y plies in the hardware, where Y=5-7 depending on X, and the Y part of the search doesn't have SE or anything like it other than things like out of check, recapture, etc extensions..." Seems understandable to me, the emails were perfectly clear and precise... > >Again... > >======================================================= > >The search occurs in parallel on two levels, >one distributed over the IBM RS/6000 SP >switching network and the other over the >Micro Channel bus inside a workstation node. >For, say, a 12-ply search, one of the workstation >nodes-working as the master for the entire >system-would search the first four plies in >software. (A ply represents a move by either >player.) > >After four plies from the current game >position, the number of positions increases >about a thousand times. All 30 workstation >nodes, including the master node, then search >these new positions in software for four more >plies. The number of positions increases by >another thousand times. > >At this point, the chess chips jump in and finish >the last four plies of the search, including quiescence >search. > >Partitioning the search into the (two-level) >software search and the hardware search per-mitted >a great deal of design flexibility, yet >maintained overall search speed. The software >handled less than one percent of the total posi-tions >searched, but it controlled about two >thirds of the search depth. The software por-tion >of the search can be arbitrarily selective >without slowing down the system. > >The eight plies of software search performed >on the RS/6000 SP included many compli-cated >search extensions, which extended the >search deeper along lines the computer con-sidered >?forcing.? Some experimental evidence >suggested that the playing strength would >increase significantly if the search extensions >went all the way down to quiescence search. > >Implementing the full software search exten-sions >on the chess chip seemed too risky a >proposition, given the design time constraint. >During the 1997 match, the software search >extended the search to about 40 plies along the >forcing lines, even though the nonextended >search reached only about 12 plies. > >DEEP BLUE IEEE MICRO > >================================================= > > >>>>Did you see the email from the DB team? Is there any misunderstanding that? >>> >>>Please post. >>> >>>Ed >>> >> >>I did, twice, in response to Vincent... > >Sorry no email of DB team seen in this thread. Perhaps you mean your own >interpretation of something nobody ever saw? OK Here is the email again: My first question to them was "did the X(Y) depth notation in Deep Thought mean X plies in software, Y additional plies in hardware as it did when I watched your program?" Here is a reply from a member of the team: Quote On--------------------------------------------------------------------- In the DT logs, the number in () after the plies is indeed the depth of the HW. It was changed dynamically, but only within a narror range (3, 4, somethimes 5). I would not be surprise if CB kept this notation, but I don't know for sure. As you say, too low, and the HW ends up idle because the host is too slow. Too high, and the host is bored... However the real reason for not letting the HW go too deep is search efficiency: the HW does not use the transposition table, hence it was best to balance the memory bandwidth availabe on the host for hash table accesses. Quote Off-------------------------------------------------------------------- The second quote came as a response to my question "Does DB report the depth using the same form of the X(Y) notation as was used in Deep Thought?" Here is the answer, again: Quote On--------------------------------------------------------------------- Hi, CB is in town this week and I had lunch with him, where we chatted a bit about DB and the like. A while back, when you looked over DB's logs (put up by Murray without IBM careing much), you were impressed by their depth and branching factor. Well, the depth notation is as I told you and just like it was in DT, so it really does go *that* deep... For example, in DT, 9(4) meant a 13 ply search. Quote Off-------------------------------------------------------------------- That is the best I can do. I asked about DT, which I saw _many_ times from behind their monitor and they explained the X(Y) notation as I have reported, several times. I then asked if DB kept the same notation and CB (nickname for Hsu) said "yes it did." Can it be any more clear than that? >Scrolling back I find this one: > >http://site2936.dellhost.com/forums/1/message.shtml?210978 > >It says 11 plies against Kasparov. It fits. They _often_ quoted the software depth. They considered the chess chips as a form of "endpoint evaluation" although they did a 5-6-7 ply search as part of that "evaluation". In Cray Blitz, I reported the depth as X, even though there were another 4 plies between the X depth and the pure q-search. These four plies didn't look at _everything_. Rather they were very selective, and included the traditional captures/checks, plus some moves that appeared to carry some sort of threat. But I said N when to be consistent with programs like Tiger of today, I could have said X+4. IE Tiger doesn't do full-width searching to depth 16. So Christophe has to explain what that 16 means when compared to others. Hsu did the same for Deep Blue... Nothing for a conspiracy theory of course... > >Your 12(6) = 18 is a fairy tale. Probably since there were very few of those. There were lots of 10(6) which I have absolutely no problem believing... and I am certain that if We were doing 10 plies at under 1M nodes per second back then, they ought to be able to go significantly deeper with 200 times (at least) the computing horsepower. At least 4 plies more not counting the futility stuff they were apparently doing in the hardware part of the search... > >Ed
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.