Author: Robert Hyatt
Date: 19:06:24 01/31/99
Go up one level in this thread
On January 31, 1999 at 16:26:24, Don Dailey wrote: >On January 31, 1999 at 10:53:57, Robert Hyatt wrote: > >>On January 31, 1999 at 04:40:11, Kim Hvarre wrote: >> >>>On January 30, 1999 at 18:15:19, Robert Hyatt wrote: >>> >>>>On January 30, 1999 at 11:38:49, Kim Hvarre wrote: >>>>>> >>>> >>>>No microcode in DB-2 at all... but it was certainly done with a 'silicon >>>>compiler' so in a sense, there is some sort of 'program', but not in the form >>>>you might think about normally... >>> >>>Isn't we around technicalities here;) >>>The basics I think is the same - microcode or "chipcoding". >>> >>> >>>> >>>>I have done that with crafty. But notice I said _match_ and not single >>>>game? That makes a difference. Also matters _where_ the game is played. >>>>IE was it just for fun, like many of the old Cray Blitz games were played? >>>>or was it a _serious_ game with something at stake to make the GM play? >>> >>>Hmm., was Kasparov playing at a serious level. Don't think so. But as You know >>>if You regularly play chess at money- or ELO-basis, it's always a matter that >>>means something - not to mention if there's a risk of getting published in front >>>of the world. >>> >>> >>>>You simply don't understand. The DB team was every bit as good as any other >>>>'team' in existance... and DB is the result of that team + time + money. Maybe >>>>Ed or others _could_ have done something (none that I know of are hardware >>>>designers which means it would be _very_ doubtful they'd have a chance). But >>>>at best, _they_ would have come up with 'deep blue'. Doubtful it would have >>>>been something "more"... >>> >>>Let's stop here. You - of all - knows there's differences between "teams" >>>(Crafty = +2300, e.g. Rebel = +2400) and the claim that the _DB-team_ is the >>>very superior, that the world at the time could establish is indeed rather >>>naive. >>> >>>regards/kim >> >> >>your statement above is _wrong_. You are making one assumption that is way >>wrong. You said 'crafty =+2300, rebel=+2400' but you forgot one important >>qualifier: 'on equal hardware'. *I* don't use 'equal hardware' and I'd be >>willing to let you fire up a test match with crafty on my box to show you what >>I mean. Or I can run it on our 16 processor SGI machine. That's the point >>here. DB's 'hardware' isn't equal. And they spent a lot of time to make it >>not equal, yet everyone overlooks that work and resorts to the lame idea of >>'if the micros had their hardware....' That's not exactly fair, is it, when >>they spent so much time to build that speed advantage, and suddenly to compare >>with them we have to strip them of that advantage? >> >>So they are as good _or better_ and their work on hardware has put them several >>levels out in front of everyone... > >The poll question was extremely ambiguious. They said something like, "if >you could run it on equal hardware ..." I made the conservative assumption >that you scale up the micro program, with no changes except adding memory >and speed to equal Deep Blue's nodes per second. > >I say this comparison is conservative because you have pointed out that >Deep Blue does so much more (because of the hardware) in the evaluation >function. In other words, comparing nodes per second is pretty unfair >to the MICRO programs if you are EQUALIZING hardware. If Deep Blue >does 1000X more nodes per second than Rebel for instance and ALSO does >100X more work for each evaluation call, then to equalize you have >to give Rebel a pentium that is 100 thousand times faster! > >Bob, I once posted that you cannot compare these things and that any >comparison is unfair and you blasted me for it. Now you are saying >exactly the same thing and blasting someone else for taking your >old point of view. You are not being fair to THEM. I hope I didn't "blast" anyone here. At least I didn't intend to do so. So I'll apologize in advance if I did. But I have _always_ taken this position I believe... that if you match DB's nps, you still need another 10x to make up for their hardware eval. I described this back when I first reported that famous 10-0 match result... If I 'blasted you' I also must have been in space warp somewhere, because I have said that same thing probably 100 times in the last 2 years... you can _not_ compare a program to DB, due to speed differences and hardware differences. So my apologies to you as well... > >Your current new point of view is in harmony with the way I feel >about it. The only fair way to compare is to put them side by >side and start playing chess. > I hope this isn't "new" for me... I wrote a detailed explanation of why this is true one post after the 10-0 result was reported, to explain that even though Hsu slowed DB down to close to a micro's speed, it still was a long way from fair because of all the free stuff they get in their hardware eval... > > > >- Don
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.