Author: Robert Hyatt
Date: 17:59:04 01/23/03
Go up one level in this thread
On January 23, 2003 at 17:56:25, Matthew White wrote: >On January 23, 2003 at 17:46:19, Robert Hyatt wrote: > >>On January 23, 2003 at 16:45:23, Gerd Isenberg wrote: >> >>>On January 22, 2003 at 16:28:47, Robert Hyatt wrote: >>> >>>>On January 22, 2003 at 12:57:40, Frank Phillips wrote: >>>> >>>>>I stumbled across a couple of interesting games by something called Suturb >>>>>against Crafty on ICC. >>>>> >>>>>There is no hardware information in its finger notes, but it seemed to be >>>>>outsearching Crafty (on dual 2.6GHz Xeon I guess), by a 1 or 2 ply at times. >>>>> >>>>>Is this the gate array chess processor thing from chessbase. >>>>> >>>>>Frank >>>> >>>> >>>>Yes. It was "brutus". It seemed to be doing about 1 ply deeper sometimes, >>>>about the >>>>same others. They claim it searches about 2.8M nodes per second in the >>>>hardware, so >>>>about 3M overall is the max, which is not a lot faster than my dual. The only >>>>thing is >>>>it uses a "Kure" book so it generally starts in a favorable position as I >>>>normally run on >>>>ICC with my "wide" book to provide variety. I would not play the same openings >>>>in >>>>(say) cct6 should I play them. :) But then again, I wouldn't play the openings >>>>I played >>>>against it against any reasonable opponent, so there you go. ;) >>> >>>Hi Bob, >>> >>>this FPGA-monster is able to do a rather sophisticated eval in parallel, so i >>>guess the "quality" of the nodes is rather huge. IMHO Chrilly's Brutus or other >>>FPGA-approaches will dominate the scene during the next years, considering that >>>FPGA hardware has much more potential for further improvement than general >>>purpose processors. More speed and more knowledge. And of course one may use >>>multiple FPGAs in some parallel framework - puh. >>> >>>Gerd >> >>I'm not sure it will really be able to dominate. For example there are already >>some hardware >>platforms that are faster overall. For example, a quad 2ghz box. My dual 2.8 >>was about as >>fast as Brutus in raw NPS, at least within reason as I was doing about 2.5M and >>they claim >>3M. I ran on a quad 2000 a month back that did just over 4M, and it won't be >>long before >>the quad 2.8's are out. >> >>yes, they can "go parallel". But the problem with any FPGA solution, is exactly >>the same >>as with the belle and deep thought/blue solutions. Hardware advancements are a >>pain to take >>advantage of, while for software solutions, we just have to wait for faster >>hardware and we are >>ready. >> >>So, I guess I am not going to fear these things any more than I feared Belle in >>1980, or >>HighTech in 1985, or deep thought in 1988. They will be tough, but far from >>invincible. >>Thank goodness. :) > >My knowledge of FPGA's is limited, but isn't their advantage that the logic >isn't "burned" in like in the other solutions? FPGA's advance in speed similar >to CPU's, don't they? > >Matt yes and no. They change internally also. And they have to plug into a PC which means that some sort of PCI bus interface is also needed. The PCI bus is changing regularly as well, as it its clock frequency. Bottom line, a real headache...
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.