Short core doubling test for SF16

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, chrisw, Rebel

Jouni
Posts: 3464
Joined: Wed Mar 08, 2006 8:15 pm
Full name: Jouni Uski

Short core doubling test for SF16

Post by Jouni »

Physical cores, 60 + 0,6, HERT book:

Code: Select all

                           
1   stockfish16 3th    +1  +10/=381/-9 50.13%  200.5/400
2   stockfish16 6th    -1  +9/=381/-10 49.88%  199.5/400

Mark Young was right :) .
Jouni
Werewolf
Posts: 1914
Joined: Thu Sep 18, 2008 10:24 pm

Re: Short core doubling test for SF16

Post by Werewolf »

Where does it say the core count?
User avatar
Eelco de Groot
Posts: 4623
Joined: Sun Mar 12, 2006 2:40 am
Full name:   Eelco de Groot

Re: Short core doubling test for SF16

Post by Eelco de Groot »

3 th is 3 threads I think, and it scores lower than 6th is 6 threads. Weird stuff. But probably result from the huge draw numbers plus statistical variation? Or, SF 16 is really not profiting from Lazy SMP under these conditions? What did Mark Young say about it Jouni ? I can't remember right.
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan
User avatar
Eelco de Groot
Posts: 4623
Joined: Sun Mar 12, 2006 2:40 am
Full name:   Eelco de Groot

Re: Short core doubling test for SF16

Post by Eelco de Groot »

Has anyone heard of Mark Young lately? He has not posted on ProDeo for months... and not here either.


By the way, this is totally off topic I know, but I have thirteen (13! it's close to 137. And a prime as well :)) Windows calculators open right now on my desktop trying to remember what I just thought up, a new conjecture about the fine structure constant, approximately 1/137, and it is really neat. Probably this message will soon be deleted because, not being about computer chess, it is totally not important. Sigh, I know, but at least someone will have read it by then :)

See also: Why Is 1/137 One of the Greatest Unsolved Problems In Physics?
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan
peter
Posts: 3272
Joined: Sat Feb 16, 2008 7:38 am
Full name: Peter Martan

Re: Short core doubling test for SF16

Post by peter »

Jouni wrote: Sun Feb 04, 2024 8:28 pm Physical cores, 60 + 0,6, HERT book:

Code: Select all

                           
1   stockfish16 3th    +1  +10/=381/-9 50.13%  200.5/400
2   stockfish16 6th    -1  +9/=381/-10 49.88%  199.5/400

Small sample size.

Code: Select all

Wins   = 10
Draws  = 381
Losses = 9
Av.Op. Elo = 3500

Result     : 200.5/400 (+10,=381,-9)
Perf.      : 50.1 %
Margins    :
 68 %      : (+  0.5,-  0.5 %) -> [ 49.6, 50.7 %]
 95 %      : (+  1.1,-  1.1 %) -> [ 49.1, 51.2 %]
 99.7 %    : (+  1.6,-  1.6 %) -> [ 48.5, 51.8 %]

Elo        : 3501
Margins    :
 68 %      : (+  4,-  4) -> [3497,3505]
 95 %      : (+  7,-  7) -> [3494,3508]
 99.7 %    : (+ 11,- 11) -> [3490,3512]
Peter.
Jouni
Posts: 3464
Joined: Wed Mar 08, 2006 8:15 pm
Full name: Jouni Uski

Re: Short core doubling test for SF16

Post by Jouni »

Yes error bar +-25. But still surprise result. With balanced book chess is basically solved.
Jouni
smatovic
Posts: 2991
Joined: Wed Mar 10, 2010 10:18 pm
Location: Hamburg, Germany
Full name: Srdja Matovic

Re: Short core doubling test for SF16

Post by smatovic »

Jouni wrote: Mon Feb 05, 2024 8:18 pm [...] With balanced book chess is basically solved.
+1

--
Srdja
User avatar
Eelco de Groot
Posts: 4623
Joined: Sun Mar 12, 2006 2:40 am
Full name:   Eelco de Groot

Re: Short core doubling test for SF16

Post by Eelco de Groot »

Eelco de Groot wrote: Mon Feb 05, 2024 3:26 am Has anyone heard of Mark Young lately? He has not posted on ProDeo for months... and not here either.


By the way, this is totally off topic I know, but I have thirteen (13! it's close to 137. And a prime as well :)) Windows calculators open right now on my desktop trying to remember what I just thought up, a new conjecture about the fine structure constant, approximately 1/137, and it is really neat. Probably this message will soon be deleted because, not being about computer chess, it is totally not important. Sigh, I know, but at least someone will have read it by then :)

See also: Why Is 1/137 One of the Greatest Unsolved Problems In Physics?
The subject matter is more difficult than I thought.... :)

Nevertheless, to not let this one totally hanging I probably will write something in the ProDeo forum if it still exists at a later date.

Still one small conjecture I postulate: if the Fine structure constant (reciprocal of) can be approximated by the latest result from https://hal.science/hal-03107990/file/main.pdf to be 137.035999206 +/- 11 in the last two decimal places, and for some reason the fine structure constant would also be some flat surface area with a simple radius, then the above result can be approximated, to a fair degree of accuracy, with the following:

π * (6 + (25 * 825)/34117)^2 = 137.03599920317158402843280124657


This should be considered as just an approximation à la Ramanujan, there is probably not a real physical meaning to considering the Fine structure constant as a two dimensional circular area... The 825 / 34117 fraction is just hypothetical, from writing a repeated fraction as a numerator divided by denominator from here: https://www.calculatorsoup.com/calculat ... tion=solve

34117 I just see, is not a prime, so you can also write it as 109 * 313.
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan