Author: martin fierz
Date: 14:07:42 09/04/02
Go up one level in this thread
On September 04, 2002 at 12:08:17, Robert Hyatt wrote: >On September 03, 2002 at 22:34:29, martin fierz wrote: > >>hi vincent, >> >>i cannot understand a word of what you say. what i see in the table is that all >>times given in the table, with the exception of the 1-processor column, are NOT >>measured numbers, but calculated numbers with the help of a factor which is >>rounded to one decimal place. >> >>>So Bob *had* to fake the outputs of 1-8 processors or his 16 processor >>>thing would look silly though it wasn't at all. >> >>these "calculated" times are there in all columns - 2,4,8,16 processors. the >>16-processor time is also calculated from a 1-decimal place factor, just like >>the 2,4,8 columns. therefore, there is nothing *in this table* to suggest that >>the times in the 16-processor column are more or less "real" than the times in >>the other columns. >> >>so.... >> >>>>times. if you invent yourself a speedup number and >>>>calculate based upon that the time, then your whole >>>>thing is a big lie simply. >>...bob explained: he had some original times, calculated the speedup, wrote that >>down, and for presentation in the paper, he didnt print the real times, but the >>original 1-processor time divided by the speedup. of course, if a student of >>mine did anything like that with his numbers i'd give him a serious talking-to, >>but if you accept the hypothesis that this is how he arrived at these numbers, >>then it is just really bad style, but not a big lie. > > >Let me again clarify. I didn't say I _did_ that to the times. I said I >definitely did that for the node counts because there was no other way to >give them. I said we _might_ have done that for the times, at the time we >were doing it for the node counts. "might have". neither of us remembers, >however, it was too long ago and it was done in the midst of other things, >such as getting ready for the 1996 and 1997 WMCCC events, and so forth. I >can't say positively about something I can't recall. even though you do not remember, the data in the table is *obviously* not really measured time. if you just divide the time for 1 processor by the time for n processors you see that immediately - all numbers come out as 1.7 or 1.9 or 7.3 or something very close like 1.703. all 2nd digits after the . come out as 0. the probability for this happening for random data is 10 to the -24... therefore, you certainly did it for the times too. >however, if I know that 16 cpus was 10.0 times faster in position X and >it took 190 seconds to get that result, then I can certainly extrapolate >the time for 1, 2, 4 and 8 processors, assuming I also know the speedups >for 2, 4 and 8 processors. And that extrapolation will be just as accurate >as the raw times, since the speedup was produced from the raw times. yes >it might lose or gain a few seconds here and there. but on a search of 2000 >seconds, 20 doesn't matter much. particularly knowing just how much the >speedup can fluctuate on the same position, even in the best of conditions. this is not the point. of course you are correct about the "it doesn't really matter" part, as long as you are not interested in knowing if your averge speedup is 1.83 or 2.0 for 2 processors. i don't understand that you are not interested in that, but that is another question... if you have an average speedup of 1.83 and report it as 2.0, then that is really stupid, and claiming more than the program really can do, and i can understand if somebody is upset by that. the error you make gets smaller as the speedup gets larger, so for the large processor numbers it really doesnt matter any more. the real point is that there is *no way* you could have measured those search times, and that if you were to claim you really did measure them, you would be a proven fraud. but, as you say, you measured the speedup to 1 digit, and not the real time, then it all makes sense - except that you did something you shouldnt really do... aloha martin > >> >>>>It is provable that all search times from 1-8 cpu's >>>>at all tests are completely not true. they are about >>>>a factor 2 too fast in order to let the 16 processor >>>>look good. >>how do you know that? if it is provable, then prove it... >>please explain that more clearly! i can definitely understand that you see a >>problem with the data in the table, but with all the rest, i don't understand >>:-( > > >I think he didn't write what he meant. Replace "too fast" by "too slow". > >If the 1 cpu test is faster than it should be, the speedup for 16 would >look _worse_. however, I didn't particularly think that 11.1 was a >particularly good result, considering it meant that over 15 million dollars >of a 60 million dollar machine was totally lost. :) > > >> >>aloha >> martin
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.