Author: Robert Hyatt
Date: 13:19:56 10/23/97
Go up one level in this thread
On October 23, 1997 at 16:17:25, Robert Hyatt wrote: >On October 23, 1997 at 15:39:05, Chris Whittington wrote: > >> >>For Bob and Bruce: >> >>Class above = 2 times. As in speed doubling every 18 months, RAM price >>halving every whenever. >> >>2 times is a class above and unacceptable IMO. 1.5 times I personally >>find too much. 1.3 I could live with. >> >>Bruce says he is just short of 2 times with his 500 alpha. Plus then the >>AMD 233's are now going to be AMD 200's. >> >>Bob can't quantify it well because of the 64 bitness. >> >>So I'll buy this: we stop arguing over 500 alphas. IMO they are still >>too much, but I'll stop arguing it. >> >>Bob and Bruce are confident they'll get 500's. They are less confident >>that the 766's will turn up. Bob and Bruce both say they don't want to >>try and get a class above, just they want to stay competitive. Don't we >>all. >> >>So, peace proposal, you guys perform on 500 alphas, leave the 766's in >>their boxes, don't use them at the tourney, don't give them out to >>others. What Dark Thought will do, Dark Thought will do, but I hope they >>can run of 500 as well. >> >>So guys, how about it, peace and moderate alphas ? > > >I don't have a firm "766". But my concern would be for someone to ask >for such a machine, get a company to jump thru several hoops to get 3 of >them assembled, tested, and shipped, with the promise that they would >get >showcased at an event that would generate a reasonable amount of >publicity, >particularly if they do well. > >This is the reason I would hate to revert to an AMD machine, period. We >asked for, and DEC responded with a machine, they are paying for >shipping, >installing software, and so forth. We'd look foolish if they visited >and >found us on an AMD. They probably wouldn't say anything, but we'd never >get another loaner from them. > >I've planned on a plain /500 from the day DEC said "yes". We'd hoped >for >a 766, but I'm still skeptical that the machine will function at that >clock >rate. IE I wonder if they have run a chess engine that beats on the >integer >functional units within the cpu, to see if eventually heat build-up >becomes >intolerable? So that was an outside shot for us anyway. > >> >>And for next year, lets discuss: >> >>a) no two types of sponsor machines as at Paderborn >> >>b) nobody more than 1.3 times the provided machines ... >> > >I think you are asking for trouble. Either make it uniform platform or >not. But 1.3 is a problem. IE how do we decide on 1.3? What benchmark >do we run? If I am getting 1.4, can't I just fake my node count by 10% >internally to make it look only 1.3x faster? > >I really believe there will be a 1ghz PPC next year. How fast you run >depends on how you are written and whether you have studied that >architecture or not. > >I'd say go uniform, period, if this is perceived as a problem. Cut the >macs out, the alphas out, the MIPS out, IBM workstations out, everyone >except for an Intel X86 at YYY mhz. If you want it to be fair, that's >the only way. Otherwise, it is simply too hard to compare machine A to >machine B, since programs vary. And using a program is a dangerous >thing >anyway as that then becomes open to skullduggery. > > >> >> >>Chris Whittington there is one other idea: we put together a "chessmark" program that we believe represents the kinds of things a chess program does. We pick a machine and run this chessmark on it. that performance becomes a "1.0". We then say bring anything you want, just it doesn't run chessmark more than 1.3X faster. That could work, because that would not penalize my 64bit design, nor someone else's vector design if they find a vector micro. It would simply measure basic machine speed... of course the compilers could influence this as well...
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.