Author: Robert Hyatt
Date: 10:10:18 02/15/03
Go up one level in this thread
On February 15, 2003 at 12:59:29, Mike Hood wrote: >On February 15, 2003 at 12:39:18, Robert Hyatt wrote: > >>On February 15, 2003 at 11:43:40, Mike Hood wrote: >> >>>A short quote from Shay Bushink's interview with Mig: >>> >>>I'm not sure our dual Athlon at home would have done much worse. We were >>>somewhat tempted to use the dual because of some initial difficulties we had >>>with the new hardware, but it was a little better on the quad. The >>>eight-processor machine wouldn't have made much of a difference, although there >>>were a few indications that it would have been a bit better in a few situations. >>> >>>http://www.chessbase.com/newsdetail.asp?newsid=799 >>> >>>Are there any users of this forum who run the deep versions of chess programs >>>(Fritz, Junior, Shredder, Crafty) on multi-processor PCs who can give opinions >>>on the performance boost they have observed? >> >> >>I have posted such numbers here many times. Crafty, on a quad at XXX mhz, will >>run anywhere from 2.5X faster to 4.0X faster than a single cpu at XXX mhz. The >>average will be around 3.0 or so. The usual "estimate" I give is this: >> >>speedup = 1 + (NCPUS -1) * 0.7 >> >>And that fits pretty well for an overall "average" speedup. >> >>However, that "formula" has only been tested to eight processors, and _most_ >>of the testing has been on 2 and 4 processors. It is not known (nor even >>expected) that this scaling will hold for larger numbers of processors. But >>for duals and quads, the performance is worthwhile. The 8-way boxes vary >>significantly in performance, depending on how they do memory. For the X86 >>8-way boxes, they simply are not very good. Other 8-way boxes (alphas, SGI >>and the like) seem to do just fine but they max out at 16-way (or maybe 32-way) >>before switching to a NUMA type approach. > >Sorry, Robert, I think I expressed my question inaccurately. I'm not interested >in the speed increase (which is easy to measure), but in the chess playing >strength increase. Is the strength increase achieved by adding a second >processor equal to, less than or greater than replacing it with a single >processor that is 1.7 times as fast? I know, it's difficult to give a >"scientific" answer without playing hundreds of games, but maybe you or someone >else can give an "intuitive" answer. Shay obviously didn't think that the quad >processor used in New York was much better than his dual processor machine at >home. That's harder to answer. The only _real_ way to answer it would be to enter two identical programs into competition with the same group of humans, and play games. One program using a single cpu, the other using a dual or quad. Personally, I believe that faster will _always_ be better. Whether you believe it is 30 rating points for going 2x faster or 100 rating points for going 2x faster is not so important. So long as it is not 0.0, then faster == better. I could run the test with Crafty on a single 550 vs crafty on a quad 550, but that would only show how speed helps crafty beat slower-crafty. It won't mean a thing about how crafty would do against an IM or GM player, or against a group of different computer opponents... However, I'd bet that going from a single to a quad (3x faster for me) is worth at _least_ 50 rating points, anywhere.
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.