Author: Tom Kerrigan
Date: 16:47:49 11/18/01
Go up one level in this thread
On November 18, 2001 at 17:45:00, Jesper Antonsson wrote: >On November 18, 2001 at 15:59:31, Tom Kerrigan wrote: >>On November 17, 2001 at 20:52:36, Jesper Antonsson wrote: >> >>>To answer the original posters question, no, there is not enough knowledge about >>>DB to create a DB-emulator. If IBM made all DB information public (which would >>>probably involve documentation work for them), I think it would take perhaps six >>>months, full time for one person, to create an emulator. >> >>I doubt that. Maybe it would take 6 months for a great implementation, but just >>hacking together a mess of eval terms does not take long. > >Sure, but you would want reasonable speed and as good emulation as possible. I I think emulation is something of a misleading word here. Perhaps reimplementation is better. All you want to do is have DB's eval terms and search rules run on a PC--you don't have to muck around with exactly how they did the RS/6000-chip interfaces and parallel searches and whatnot to get something that plays like DB. >Did I say one should implement the parallel parts? No, but if you want to >replicate behaviour, then you have to consider what the parallelism in the >original *does*. You could try to do a cycle-level simulation of the entire DB system to replicate the race conditions and whatnot that they would have encountered, but what's the point? We're never going to know that level of detail about DB, so it would be impossible anyway, and why try to replicate problems that the DB encountered and solved that are specific to their setup? >I said "probably", and I base it on things Bob said years ago. But think about >it, if you could do complex eval for free in hardware, would you only implement >stuff that could be efficiently implemented in software too? I wouldn't, so I >expect that at least some of the stuff they did would run slowly in software. Ahhhh, Bob. Ask yourself why complex eval on these ASICs would be "free"... I agree that based even on the DB literature we have now, it definitely has some terms that would run slowly in software. But hey, so does MChess. That's my point. -Tom
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.