Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: SMP variability some real data to look at (now three 2cpu runs)

Author: Rolf Tueschen

Date: 04:49:16 09/06/02

Go up one level in this thread


On September 06, 2002 at 00:29:02, Robert Hyatt wrote:

>pos        1cpu       2cpu          2cpu          2cpu
> 1         153      89 (1.72)     88 (1.74)     90 (1.70)
> 2         139      90 (1.54)     90 (1.54)     90 (1.54)
> 3         130      82 (1.59)     87 (1.49)     87 (1.49)
> 4         176     100 (1.76)     99 (1.78)    100 (1.76)
> 5         147      95 (1.55)     99 (1.48)     87 (1.69)
> 6         135      77 (1.75)     77 (1.75)     78 (1.73)
> 7          92      50 (1.84)     50 (1.84)     50 (1.84)
> 8         149     104 (1.43)     89 (1.67)     90 (1.66)
> 9          80      60 (1.33)     62 (1.29)     60 (1.33)
>10         155      82 (1.89)     76 (2.04)     75 (2.07)
>11         142      80 (1.77)     80 (1.77)     82 (1.73)
>12         105      76 (1.38)     76 (1.38)     76 (1.38)
>13         149     105 (1.42)    105 (1.42)    107 (1.39)
>14         147     104 (1.41)     87 (1.69)     74 (1.99)
>15         150      86 (1.74)     86 (1.74)     86 (1.74)
>16         159      73 (2.18)     72 (2.21)     72 (2.21)
>17         158      97 (1.63)     97 (1.63)     77 (2.05)
>18          85      42 (2.02)     41 (2.07)     42 (2.02)
>19         125      61 (2.05)     63 (1.98)    103 (1.21)

How should we interprete just in normal language or chess this result? Two times
almost factor 2 and then down to 1.21?

Rolf Tueschen

>20         160     112 (1.43)    117 (1.37)    109 (1.47)
>21         156      73 (2.14)     62 (2.52)    103 (1.51)
>22          77      60 (1.28)     73 (1.05)     60 (1.28)
>23         134      93 (1.44)     95 (1.41)     97 (1.38)
>24         131      84 (1.56)     82 (1.60)    115 (1.14)
>  average speedup->    (1.66)        (1.69)        (1.72)
>
>
>OK... some more food for thought.
>
>1.  The overall average speedup is "smoother" than I thought.  I am going to
>add the other computational approach to show the average of the 24 speedups,
>and then the total speedup computed by dividing the sums of the times in the
>columns...
>
>2.  There is a interesting variability.  A few positions are pretty much
>"rock solid" in their times.  Like #18, and a few others.  A few are
>all over the spectrum, #24 is one, but others like #14 are really bouncing
>around too.
>
>3.  Each run produces 2-3-4 super-linear speedups, and a couple of
>positions are consistent.  That causes me some concern that move ordering
>is having problems there and I am interested in the "why" of that...
>
>4. The fail high percent for a wild position seems normal (91% for #14 for
>example).
>
>5.  the fail high percent for the superlinear positions also seems sane.  But
>since a couple of positions do it every time, I'm going to look at some partial
>trees to see what is up with that.  Although super-linear is not unexpected, I
>am seeing more than I did with CB.  Perhaps the fact that CB's evaluation didn't
>produce these enormous positional scores kept the tree shape more under control,
>I don't know (yet).
>
>6.  It looks like the 1.7X speedup for 2 cpus is pretty solid here as well,
>as I had seen in the past...
>
>I have one more of these running, when it is done I will produce one final
>table and then maybe queue up 4 of the 3cpu tests and do it again.  I am
>pretty sure the variability will climb with the number of processors...
>
>If anyone wants the large raw data logs, let me know, I will save them for a
>while.  If anyone wants to check my eyeball copying of the data, (the times)
>that would certainly be good, but you will need the raw logs.



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.