Sedat Canbaz

[Sedat Canbaz]

Btw, sure without running this test...I can not...

So far, (after 137 games out of 500 games...)
Raubfish SC NNUE leads 250+ Elo over Cfish 030820 NON-NNUE:

Code: Select all

Score of Cfish NON-NN vs Raubfish SC NNUE: 0 - 86 - 51 [0.186]
...      Cfish NON-NN playing White: 0 - 36 - 34  [0.243] 70
...      Cfish NON-NN playing Black: 0 - 50 - 17  [0.127] 67
...      White vs Black: 50 - 36 - 51  [0.551] 137
Elo difference: -256.3 +/- 47.0, LOS: 0.0 %, DrawRatio: 37.2 %
SPRT: llr 0 (0.0%), lbound -inf, ubound inf
137 of 500 games finished.

Note: Both players are with 2m+1s, 128 mb hash, Balsa openings etc.

[Sedat Canbaz]

Update

After 354 games, 234+ Elo in favor for Raubfish SC:

Code: Select all

Score of Cfish NON-NN vs Raubfish SC NNUE: 1 - 209 - 144 [0.206]
...      Cfish NON-NN playing White: 1 - 83 - 93  [0.268] 177
...      Cfish NON-NN playing Black: 0 - 126 - 51  [0.144] 177
...      White vs Black: 127 - 83 - 144  [0.562] 354
Elo difference: -234.2 +/- 27.6, LOS: 0.0 %, DrawRatio: 40.7 %
SPRT: llr 0 (0.0%), lbound -inf, ubound inf
354 of 500 games finished.

That's all for today.. because Turkey vs Croatia Football match is started

Greetings

[Sedat Canbaz]

New Update

A lot of book, engine, hash etc. testings are added, updated:
https://sites.google.com/site/computers ... testings-5

GAMES:
https://mega.nz/file/2phVVSYR#OcXy4JiW2 ... Ypmbr7q2Jg

Btw, right now a new test is started: DON NON-NN (15m+1s) vs
Top books NN (30s+0.5s) and later we will have more stats for
Deep Books+Bullet+NN is close in strength to Rapid NON-NN ?!
And please stay tuned...more updates very soon...


Best,
Sedat

[Sedat Canbaz]

Another Update

LIVE ranking: DON NON-NN (15m+1s) vs Top Deep Books NN (30s+0.5s)

Sure, too early for any conclusion...bıt under current cond..
Fast Bullet NN seems to be more close to Rapid NON-NN
Ok.. after 550 games will will have more serious data..

Note also that (regarding normal engine testings, not for book tours),
Even 500 games (per player) can not be considered as very serious data..
E.g 1000+ games seems to be ideal...in case of testings via strong books!
But no doubt that matches like 100 or 200 or 300 games..the error margin
May lead to wrong conclusions..at least many many tests indicate this..!!
Just in case of weak openings, then at least 3000 games (per player) required..
How to know which opening suite is weak...usually the ones, which produce
Below than 80%-85% draw ratio... and hope helps... !!

Code: Select all

Rank Name                          Elo     +/-   Games   Score    Draw 
   0 DON NON-NN                     12      22     173   51.7%   82.7% 
   1 ORIONPAX3                      95      86      15   63.3%   73.3% 
   2 ShadRach                       29     100      12   54.2%   75.0% 
   3 Red Dragon                     25      47      14   53.6%   92.9% 
   4 Zaytsev                        20      70      17   52.9%   82.4% 
   5 SENTINEL 2                      0      57      17   50.0%   88.2% 
   6 OPTIMUS 4                       0       0      16   50.0%  100.0% 
   7 OlisFish 260923                 0       0      17   50.0%  100.0% 
   8 CachoSCCT opt                   0      93      15   50.0%   73.3% 
   9 Everest 300923                -22      42      16   46.9%   93.8% 
  10 Pimienta                      -70      74      15   40.0%   80.0% 
  11 LesChats 071023              -179     106      19   26.3%   52.6% 

173 of 550 games finished.

[Sedat Canbaz]

New Update:

DON NON-NN (15mim+1sec) vs Top deep books NN (30sec+0.5sec):
https://sites.google.com/site/computers ... testings-5

It seems via DON Hercules 2 book:
Deep Books+Bullet+NN is close in strength to Rapid NON-NN

Btw, ORIONPAX3 book is doing very good, congrats to Mr. Roger!!

Code: Select all

Rank Name              Elo    +    - games score oppo. draws 
   1 ORIONPAX3         3778    8    8   700   53%  3758   88% 
   2 DON NON-NN 15m1s  3772    9    9   550   51%  3762   82% 
   3 DON Hercules 2    3769    9    9   550   51%  3762   90% 
   4 Everest 300923    3769    8    8   700   51%  3759   83% 
   5 CachoSCCT opt     3766    8    8   700   51%  3759   89% 
   6 ShadRach          3766    8    8   700   51%  3759   87% 
   7 OlisFish 260923   3763    8    8   700   51%  3760   94% 
   8 SENTINEL 2        3763    8    8   700   51%  3760   91% 
   9 OPTIMUS 4         3761    8    8   700   50%  3760   93% 
  10 LesChats 071023   3760    8    8   700   50%  3760   86% 
  11 Zaytsev           3755    8    8   700   49%  3760   90% 
  12 Pimienta          3755    8    8   700   49%  3760   86% 
  13 Red Dragon        3749    8    8   700   48%  3761   87% 
  14 DON NON-NN 8m1s   3742    9    9   550   47%  3762   81% 
  15 DON NON-NN 2m1s   3729    9    8   550   45%  3762   80% 

And please stay tuned...more/new updates very soon...e.g as
Next Duel: Balsa book+Bullet+NN vs Balsa book+Rapid+NON-NN

Greetings

[Sedat Canbaz]

New Update

4th test: Balsa openings, 30s+0.5s NN vs 15m+1s NON-NN
E.g according to current test: +57 Elo (for Bullet NN)

Code: Select all

1 Raubfish X48c3 SC NNUE  +125/-43/=332 58.20%  291.0/500
2 Cfish 030820 NON-NN     +43/-125/=332 41.80%  209.0/500

Games:
https://mega.nz/file/D9g1CZqJ#WXl0KP53e ... vfK6OQFoIg

To be more clear,
Via Balsa, 30s+05s+NN seems to be clearly stronger than 15m+1s
NON-NN e.g according to recent above testings approx. +57 Elo
And why via Balsa openings more Elo diff. that DON's openings?
Simply because DON's openings are more deeper than Balsa...
In same time that means (going deeper) NON-NN gains more time!
In short, giving more time means more eng strength, Elo points..

What I can say more,
Just imagine before NNUE era: Rapid TC were considered as a good quality..
At present, via NN even 30s+0.5s can be on same level as in the past-Rapid !
After all I think that the real problem is in our brains...if Bullet bad.. ))
It is time for new revolution !)

Ok...that's all for now... see you later )

For more details:
https://sites.google.com/site/computers ... testings-5


Best,
Sedat

[Sedat Canbaz]

Hello Friends,

I am pleased to announce that,
I started another new testing...

In short,
Let's see after 800 games (per player)
Will there be significant improvement or not..
Between the current official engine releases?!

As alternative, the latest:
Stockfish 101023 dev is participating as well...

And here is current LIVE ranking:

Code: Select all

Rank Name                          Elo     +/-   Games   Score    Draw 
   1 Stockfish 101023               25      14     152   53.6%   92.8% 
   2 Stockfish 16                   16      12     149   52.3%   95.3% 
   3 Stockfish 15.1                 -5      13     150   49.3%   94.7% 
   4 K.Dragon 3.3                  -10      13     146   48.6%   94.5% 
   5 K.Dragon 3.2                  -29      17     143   45.8%   90.2% 

370 of 2000 games finished.

Conditions:
2x Epyc 7B12, HT+LP OFF, CuteChess, 1 Core, PonderOFF,
Balsa opening suite, 1min+0.6sec, 128 MB Hash, 4-MEN

More info very soon..

Good Luck!

[Sedat Canbaz]

Update 2

Code: Select all

Rank Name                          Elo     +/-   Games   Score    Draw 
   1 Stockfish 101023               22       9     484   53.2%   91.9% 
   2 Stockfish 16                   17       7     481   52.4%   94.4% 
   3 Stockfish 15.1                 -2       8     479   49.7%   93.5% 
   4 K.Dragon 3.3                  -10       8     473   48.5%   93.2% 
   5 K.Dragon 3.2                  -27      10     475   46.1%   89.3% 

1196 of 2000 games finished.

[Sedat Canbaz]

Update 3

Bullet competition is ended...and as I expected,
Under these cond...no any SF Elo improvement!
Sure I am referring during latest past 4 months...

Sure I am not as engine programmer, very likely
The experts will explain better than Me...but it
Seems hard to be improved via strong openings!?

But the good news is that (since last year 2022),
KomodoDragon 3.3 is gained +10 Elo over v3.2, nice !
Where SF16 improved +15 Elo over v15.1, not bad too!

Code: Select all

Rank Name                   Elo   +/-  Games  Score    Draw 
   1 Stockfish 101023       17     7    800   52.5%   92.0% 
   2 Stockfish 16           17     6    800   52.4%   93.5% 
   3 Stockfish 15.1          2     6    800   50.3%   93.9% 
   4 K.Dragon 3.3           13     7    800   48.1%   92.1% 
   5 K.Dragon 3.2          -23     7    800   46.8%   90.5% 

2000 of 2000 games finished.

Note also that,
I started with same players a Blitz (3min+1sec) tourney,
So I hope to share all the games after the end of tours...

And please stay tuned...more updates very soon !

[Sedat Canbaz]

Blitz Update 1

Code: Select all

Rank Name                          Elo     +/-   Games   Score    Draw 
   1 Stockfish 16                   13       9     319   51.9%   94.4% 
   2 Stockfish 101023               10       7     322   51.4%   96.6% 
   3 Stockfish 15.1                  4       7     318   50.6%   96.9% 
   4 K.Dragon 3.3                  -11       8     315   48.4%   95.6% 
   5 K.Dragon 3.2                  -17       8     312   47.6%   95.2% 

793 of 1500 games finished.