Author: Slater Wold
Date: 15:56:33 09/01/02
Go up one level in this thread
On September 01, 2002 at 17:54:13, Alexander Kure wrote: >On September 01, 2002 at 17:37:54, Slater Wold wrote: > >>On September 01, 2002 at 15:38:35, Alexander Kure wrote: >> >>>Dear Slater, >>> >>>>2 - 3 million is still somewhat slow. If you take into consideration what HW >>>>will be used at the next WCCC, it is not hard to imagine Crafty will be getting >>>>close to 3M nps, and Deep Junior close to 6M. >>>> >>>>So you're back to relying on a higher quality search. >>>> >>>>However, I understand this can be remedied with a better FPGA or even more >>>>FPGAs. But of course, there is a reason why you're not doing that now. $. >>> >>>2 - 3 million is excellent if you consider the complex evaluation which is done >>>in the FPGA! The FPGA does not suffer from speed loss when adding more knowledge >>>to the evaluation function because of paralellism! Brutus is not considered to >>>be a fast searcher but a high class evaluator. >>> >>>Greetings >>>Alex >> >>That is, in all my reading, not what HW evals are about. >> >>Of course 'knowledge' doesn't cost in the form of NPS, but it does cost in the >>form of gates. And that's not something you have an unlimited supply of. > >Right, but we still have plenty of gates available ;-) Really? From what I understood, you were using a relativley small FPGA. Would there be anyway we could possibly get a "synopsis" of Brutus? Such as the host machine, what kind of FPGA it is running. What kind of eval (and how big it is in gates), so on and so forth. That to me, would be outstanding. I understand it's not a finished product, but like I said, maybe just a short summary of what it is, and how it is running.
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.