Genius isn't sensitive enough?
Serge Desmarais
I was wondering about something that could, maybe, explain why Genius
dropped from the top of the SSDF's list. Maybe it has to do with a lack
of CPU increase sensitivity?
I was looking at the SSDF list and compared the rating of several
programs running on a 486/DX2-66 and the same program running on a
Pentium 90. Here are the results :
-Genius 4.0 went from 2379 to 2411, an increase of 32 points, so +1.3%.
-MChess Pro 4.0 went from 2314 to 2360, +46 ELO points and +2.0%.
-Genius 3.0 went from 2358 to 2410, gaining 52 points (+2.2%).
-MChess Pro 5.0 gained 69 points (from 2342 to 2411), +2.9%.
-Rebel 7.0 was better, still, from 2335 to 2408 (+73 points, +3.1%).
-HIARCS 3.0 was 2300 on a 486, and is 2379 on a Pentium (+79 points and
+3.4%).
-Fritz 3.0 gains 98 points, from 2256 to 2354! (+4.3%).
-Rebel 6.0 is the champion, from 2300 to 2412 (a gain of 1112 points!! -
+4.9%!).
And if U look at the results of the matches between these programs and
themselves, but running on slower machines, U see that :
-Rebel 6.0 BLASTED itself 16 to 4! (it just lost 9.5 to 10.5 against
Rebel 8!).
-Fritz 3.0 CRUSHED itself 15.5 to 4.4, no less!
-HIARCS 3.0 won easily 12.5 to 7.5.
-Rebel 7.0 did the same, winning 14.5 to 7.5.
-MChess Pro 5.0 DREW 10-10!!!
-Genius 3.0 LOST 9 to 11!
-MChess Pro 4.0 won 12 to 8.
-Genius 4.0 won 12.5 to 7.5 too.
So, though Genius 3, for example, is 102 points above Fritz 3.0 when
both are on a 486, Genius is just 56 points above on a Pentium 90.
Question : on WHAT machine and up would Fritz 3.0 pass AHEAD of Genius
3.0, 4.0 and 5.0? On WHAT machine and up would Rebel 6.6 become #1 on the
list? On the Rebel web page, they are reporting testing matches abd how
their Rebel 8's winning percentage increases with the CPU... But they
match it with Genius! Now, one could make the hypothesis that maybe Fritz
3.0 could win if both were on a Pentium MMX/200? Or Rebel 6.0 that lost
just by one point to Rebel 8.0 on a Pentium 90, could tie the match on a
Pentium 100, win by a small margin on a Pentium 120, win with a
perceptible advance on a Pentium 133, blast Rebel 8.0 on a Pentium 150
and crush Rebel 8 like a bug on a Pentium 200? Somewhat kidding, but
still, one may wonder!
My point is that every new Genius program is better than the
preceeding ones, as opposed to what Lonnie says that Genius 5.0 is weaker
than the 2 preceeding ones, but when U put them on a faster machine, that
doesn't make them significantly better... WHY? The code is too slow? It
is too much of a "knowledge" program? So, as the SSDF is testing the
programs with faster and faster materials, Genius would be far and far
behind?
I am curious to know what U think about all these... :)
Serge Desmarais
Robert Hyatt
Serge Desmarais (psy...@total.net) wrote:
: I was wondering about something that could, maybe, explain why Genius
these are typical. everyone sees that even a small change in *anything* can make
self-play produce wildly skewed results. Obviously 16-4 is better, but 200+ rating
points? Hardly. computerx vs computerx produces such remarkable disparities.
Computerx vs computery is better, but still can produce skewed results...
: So, though Genius 3, for example, is 102 points above Fritz 3.0 when
mclane
Serge Desmarais <psy...@total.net> wrote:
> My point is that every new Genius program is better than the
>preceeding ones, as opposed to what Lonnie says that Genius 5.0 is weaker
>than the 2 preceeding ones, but when U put them on a faster machine, that
>doesn't make them significantly better... WHY? The code is too slow? It
>is too much of a "knowledge" program? So, as the SSDF is testing the
>programs with faster and faster materials, Genius would be far and far
>behind?
>I am curious to know what U think about all these... :)
You are wrong. Genius3 is stronger Genius4 is stronger Genius5.
Genius3-4-5 are all in assembler 16 Bit code!
I don't think that Richards program has made ANY progress.
Genius4 sometimes overstresses the position, this leads to more active
play, but also to more losses.
Genius5 is vice versa. But not better than Genius3.
That is - of course - only my opinion.
>Serge Desmarais