Author: Matthew Hull
Date: 10:55:18 05/21/04
Go up one level in this thread
On May 21, 2004 at 13:08:54, Vincent Diepeveen wrote: >On May 21, 2004 at 12:58:19, José Carlos wrote: > >>On May 21, 2004 at 12:24:12, Vincent Diepeveen wrote: >> >>>On May 21, 2004 at 12:22:36, Matthew Hull wrote: >>> >>>>On May 21, 2004 at 12:13:09, Vincent Diepeveen wrote: >>>> >>>>>On May 20, 2004 at 22:41:23, enrico carrisco wrote: >>>>> >>>>>>On May 19, 2004 at 22:24:08, Vincent Diepeveen wrote: >>>>>> >>>>>>>On May 19, 2004 at 12:18:04, Robert Hyatt wrote: >>>>>>> >>>>>>>>On May 19, 2004 at 10:29:28, Vincent Diepeveen wrote: >>>>>>>> >>>>>>>>>On May 18, 2004 at 14:07:37, Robert Hyatt wrote: >>>>>>>>> >>>>>>>>>>On May 18, 2004 at 13:52:29, Vincent Diepeveen wrote: >>>>>>>>>> >>>>>>>>>>>On May 18, 2004 at 13:25:18, Robert Hyatt wrote: >>>>>>>>>>> >>>>>>>>>>>>On May 18, 2004 at 12:34:31, Vincent Diepeveen wrote: >>>>>>>>>>>> >>>>>>>>>>>>>On May 18, 2004 at 11:44:27, Robert Hyatt wrote: >>>>>>>>>>>>> >>>>>>>>>>>>>Yes, you can't afford to leave USA 1 day, but you can afford $15k+ >>>>machines >>>>>>>>>always. >>>>>>>>>>>> >>>>>>>>>>>>I don't own a single 15K machine, period. I own one sony laptop, one >>>>gateway PC >>>>>>>>in my home. >>>>>>>>>>>> >>>>>>>>>>>>And you talk about "selective math". In your case it is "non-math" as >>>>every >>>>>>>>number you puke up is utter nonsense. >>>>>>>>>>> >>>>>>>>>>>So you deny that you wrote speedup = 8.81 in your thesis >>>>>>>>>>>and that you wrote in your DTS article speedup = 11.1 >>>>>>>>>> >>>>>>>>>>Please quote where I denied that. I didn't deny _either_ result... >>>>>>>>> >>>>>>>>>8.81 != 11.1 >>>>>>>>> >>>>>>>>>and your 11.1 results are based upon data which can be proven as a big >>>>fraud. >>>> >>>>>>>> >>>>>>>>First, 8.81 came from BK at 5 plies. 11.1 came from a set of game >>>>positions at >>>>10 plies. 8.81 carried nothing from position to position. >>>>11.1 carried >>> >>>>>>>Your 11.1 comes from nowhere. You invented it yourself. Based upon self >>>>invented >>>speedup numbers you calculated then search time. This is trivial to >>>>proof and >>>has been proven in 2002 august. >>>>>> >>>>>>Okay. If it is so trivial -- please "re-prove" it. I missed this August >>>>2002 >>discussion of proof. I would, however, be interested in seeing the >>>>proof >>(rather than more threads of rants, raves, and name calling...) >>>>> >>>>> >>>>>pos 2 4 8 16 >>>>>1 2.0000 3.40 6.50 9.09 >>>>>2 2.00 3.60 6.50 10.39 >>>>>3 2.0000 3.70 7.01 13.69 >>>>>4 2.0000 3.90 6.61 11.09 >>>>>5 2.0000 3.6000 6.51 8.98876 >>>>>6 2.0000 3.70 6.40 9.50000 >>>>>7 1.90 3.60 6.91 10.096 >>>>>8 2.000 3.700 7.00 10.6985 >>>>>9 2.0000 3.60 6.20 9.8994975 = 9.90 >>>>>10 2.000 3.80 7.300 13.000000000000000 >>>>>and so on for all positions >>>>> >>>>>That is the speedup from Hyatt if i simply divide the searchtimes through >>>>single >cpu times. >>>>> >>>>>If diep however runs parallel i get not such nice numbers. >>>>> >>>>>For 2 processors i could get for example: 2.39 1.34 1.01 2.02 1.93 and so on. >>>> >>>> >>>> >>>>Is it not true that DIEP uses a different parallel strategy than crafty or Cray >>>>Blitz? If so, then one should not be surprised that their results are not the >>>>same. >>>> >>>>So you are comparing apples and oranges, yes? >>>> >>>>Most programmers can understand this. >>>> >>> >>>LEARN TO READ what i write. What is your IQ? >>> >>>What i write is that his search times cannot be original search times but that >>>they are frauded search times. >>> >>>You do not get search times that go to perfect speedup numbers for so many >>>speedup numbers. >> >> >> This has all been explained in the past (see the archives). >> For a quick analogy (for those who don't want to read a million posts): >>Suppose you compute your nps for 100 positions, and save the numbers. Then you >>are asked for the exact time. You don't have it, but you have the node count. >>You just multiply node count by nps and give an approximate time. Not exactly >>the same, but same idea. >> No fraud. Just too difficult for Vincent to understand. But he'll get it when >>he grows up, of course. >> >> José C. > >The difference is, this article is the only article in history and the only data >in history in an official journal that proves the dts algorithm as implemented >by hyatt to get 11.1 speedup at 16 processors. > >His thesis says 8.81 > >Do you get it? The 11.1 and 8.81 were done at different search depths on different acrchitectures and for different sets of positions (IIRC). That I did get, which means your accusation is spurious. > >If i get a complete harddisk crash here, the logfiles from diep world champs >2003 at a 512 processor machine are still in possession by several other chess >program authors. > >And i never wrote an official article about it! > >> >>>> >>>>> >>>>>Not nice distributed numbers. >>>>> >>>>>I have it in excel available (thanks to Ron Langeveld) >>>>> >>>>>>-elc. >>>>>> >>>>>><snipped>
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.