Page 9 of 63

Re: Sedat Canbaz

Posted: Fri Jul 28, 2023 9:27 am
by massimilianogoi
Sedat Canbaz wrote: Thu Jul 27, 2023 8:18 am Hello Friends )
...
Actually Dual processors are cooled via 2x NH-U12S TR4-SP3, but against
Any kind of risks...for these testings, Air Conditioner is used as well )
You may know, Dual EPYC machine is not so cheap, at least where I live!

As Final Words,
I'm doing my best since many years..mainly for Computer Chess's progress!
And I hope that you like my chess activities.. sure they are not perfect, but..
That's all from Me... !)
...
They are significant for what we do! I just took a peak on that cooler model... Aren't we sure it wouldn't be better - in your specific case (huge load of work for a long time) - a liquid cooler? If I were you I would be afraid as hell for my computers with just air cooling...

Re: Sedat Canbaz

Posted: Fri Jul 28, 2023 9:30 am
by massimilianogoi
Sedat Canbaz wrote: Fri Jul 28, 2023 4:27 am Hello Friends,

Meanwhile,
In next 2-3 weeks, I am planning to be on Holiday..

So.. please do not worry much...I am all right )

Sure later when I'll have more free time ...
I will try to be active again..at least on forums )

Best,
Sedat
Have a good time! Image

Re: Sedat Canbaz

Posted: Fri Jul 28, 2023 11:23 am
by Sedat Canbaz
massimilianogoi wrote: Fri Jul 28, 2023 9:27 am
Sedat Canbaz wrote: Thu Jul 27, 2023 8:18 am Hello Friends )
...
Actually Dual processors are cooled via 2x NH-U12S TR4-SP3, but against
Any kind of risks...for these testings, Air Conditioner is used as well )
You may know, Dual EPYC machine is not so cheap, at least where I live!

As Final Words,
I'm doing my best since many years..mainly for Computer Chess's progress!
And I hope that you like my chess activities.. sure they are not perfect, but..
That's all from Me... !)
...
They are significant for what we do! I just took a peak on that cooler model... Aren't we sure it wouldn't be better - in your specific case (huge load of work for a long time) - a liquid cooler? If I were you I would be afraid as hell for my computers with just air cooling...
Well, on my older i7 980X machine
I had liquid cooling system but later I switched to air cooling...reason after some period of time.. liquid system started to give troubles...

On other hand
Even if I try to build a custom luquid cooling system...then I am afraid to say that
No any liquid cooling parts to be available on market...I mean to be suitable for 2x Epyc 7B12 ..

Forget everything
Even no much choices to find air coolers for these kind of processors

What I can say more
Over the past years.... My machines are cooled with air coolers...but latest days the weather temps are too high..minus plus 40 - 45 degrees...
So for these reason...I have to be more careful )


Greetings

Re: Sedat Canbaz

Posted: Sun Jul 30, 2023 7:44 am
by Sedat Canbaz
Hello there )

Well, before taking a good rest..I realized to share some
Data: based since 2021, but mostly belongs to 2022 / 2023

To be more clear,
Here are useful competitions, e.g now we can compare
SCCT's learning engines as best performance statistics!

So it is clear that,
Their eng learning effect played as serious role, otherwise
I hardly doubt they'd get so high points and to be Winners!
Plus as other PROOF, we noticed in latest recent testings,
E.g Eman and RapTora are started to perform better in next
Re-run testings...and very likely the learned exp helped...

Eman 8.91: Winner of SCCT (2m+1s) Duel - Eman Vs RaptoRa:
https://sites.google.com/site/computers ... vs-raprora

Eman 8.91: Winner of SCCT (2m+1s) NNUE Championship XXXIV:
https://sites.google.com/site/computers ... t-nn-cs-34

RapTora 2.2: Winner of SCCT (1m+0.6s) NNUE Championship XXXIII:
https://sites.google.com/site/computers ... t-nn-cs-33

RapTora 2.2: Winner of SCCT (1m+1s) NNUE Championship XXXII:
https://sites.google.com/site/computers ... t-nn-cs-32

SugaR AI SE: Winner of SCCT (2m+1s) NNUE Championship XXXI:
https://sites.google.com/site/computers ... t-nn-cs-31

RapTora 2.2: Winner of SCCT (2m+1s) DUEL-Book vs Experience III:
https://sites.google.com/site/computers ... k-vs-exp-3

StockfishMZ 170522: Winner of SCCT (2m+1s) NON-NNUE CS VI:
https://sites.google.com/site/computers ... on-nn-cs-6

Raubfisch X48e: Winner of SCCT (2m+1s) NNUE Championship XXI:
https://sites.google.com/site/computers ... nnue-cs-21

Eman 6.91+SugaR AI 1.40: Winners of SCCT (3m+2s) Handicapped CS II:
https://sites.google.com/site/computers ... apped-cs-2

Eman 6.91: Winner of SCCT (3 min) NNUE Championship IV:
https://sites.google.com/site/computers ... -nnue-cs-4

By the way, I still believe in that,
A super strong exp file (well-prepared, much trained vs books),
Can be as Champ or if nothing else to be as one of Top ranks at:
https://sites.google.com/site/computers ... t-ben-cs-1

And meanwhile, for example:
RapTora exp managed to be Winner vs Books (in Duel III...)
And in next time/s, why not other/s ?)
If at first you don't succeed, try again and again!
A failure is just another step.. closer to success!
But before, I suggest harder work...otherwise no BIG
Chances... because the current Top books are really
Very dangerous...proved many times in older tours!


Best,
Sedat

Re: Sedat Canbaz

Posted: Thu Aug 03, 2023 10:12 pm
by Sedat Canbaz
Hello Chess Friends )

Well, while we are testing Top CPU chess engines,
A new idea is born:
Let Me say please 'my view' about the BEST ones!

1st of all, in my opinion:
Best of Best is Stockfish !!

And why? if not clear, simply because without SF:
- Many of the Top chess engines would not exist!
- Plus many would not be so strong as nowadays!

And according to overall SCCT's data:
Here are some of the rest BEST ones:

Best Learning 3700+ Elo Engines:

1. RapTora - Super strong, stable, drawish, BIG size
2. Raubfisch SC - Strong, stable, LESS drawish than all, small size
3. LittleBeast - Super strong, stable, drawish, small size
4. BrainLearn - Super strong, stable, drawish, very BIG size
5. Eman - Super strong, NOT stable, drawish, BIG size


------------------------------------------------------------------------------

Best NON-Learning 3700+ Elo Engines:

1. SF-PB - Super strong, stable, drawish, small size
2. Kayra - Super strong, stable, drawish, small size
3. SF Polyglot - Super strong, stable, drawish, BIG size
4. Kookaburra - Super strong, stable, drawish, BIG size
5. KomodoDragon - Strong, stable, LESS drawish, BIG size


More Details,
Note that only one chess engine (per author) is allowed
I've picked mainly SCCT's Winners (played in 2022/2023)
Plus as tester, from my point of view as most important:
Strength, Stability, Size, sure which are less Drawish...
Unfortunately M.Z's engines are banned, so not allowed...
As other important factor: Professionalism required too!
And only to be 'good' in engine programming: NOT enough!

On other hand, if any other engine is missed, I'm sorry..
Btw, LcO is one of best.. but requires fast video card..
Cfish is great one too, but no free time to list all...etc.
And frankly there are many other Top engines, but as far
As possible I payed attention to public engines of 3700+
Exc. Eman and Raubfisch, but without them I could not...
As reason, many times Eman managed to win, where
Raubfish SC less drawish than all+ most frequent engine!

Ok...that's all for now..and thanks for your interest !)


Best,
Sedat

Re: Sedat Canbaz

Posted: Sat Aug 05, 2023 2:26 pm
by Sedat Canbaz
Hello Friends,

Well, here is a new Champ stats based on SCCT Ratings,
Which are played in the latest in 2-3 years (2020- 2023):

Code: Select all

Champion/Winner     Engine Author         Championship
Eman 8.91           Khalid Omar/SF Team     NNUE XXXIV
RapTora 2.2         Anton Ces/SF Team      NNUE XXXIII
RapTora 2.2         Anton Ces/SF Team       NNUE XXXII
Polyfish/SugaR      K. Omar M. Zerbinati     NNUE XXXI
Stockfish 220423    SF Team                   NNUE XXX
Stockfish 220423    SF Team                  NNUE XXIX
Stockfish 220423    SF Team                NNUE XXVIII
Stockfish 220423    SF Team                 NNUE XXVII
Stockfish 220423    SF Team                NNUE XXVIIb
RapTora 2.2         Anton Ces/SF Team  Book Vs Exp III
SF-MZ/SF SkyNet     M.Z/SkyNet/SF team     NON-NNUE VI
Cfish 030820        Cfish/SF Team         NON-NNUE VIb
Stockfish 15.1      SF Team                   NNUE XXV
Kookaburra 1.03a    K./SF Team               NNUE XXIV
FisherovMonk        Fisherov/SF Team        NNUE XXIII
Kayra 1.6           M.Karaman/SF Team        NNUE XXII
SF-PB/Zeus          Alex/Zeus/SF Team      NNUE XXIIb
Raubfisch X48e      Raubfisch/SF Team         NNUE XXI
Stockfish 15        SF Team                    NNUE XX
Kayra/Kookaburra/SF Polyglot/SF-PB            NNUE XIX
SF-PB 130722        Alex/SF Team     Handicapped/NN I
Cfish 030820        Cfish/SF Team           NON-NNUE V
Polyfish 220826     Khalid Omar/SF Team     NNUE XVIII
Stockfish 170922    SF Team                  NNUE XVII
Stockfish 110922    SF Team                   NNUE XVI
Carlsen + Stockfish SF Team                 Human Like
Cfish 030820        Cfish/SF Team          3600+ Elo V
Kayra/SF200620MZ/SF Polyglot              3600+ Elo IV
SF-PB/SF Polyglot   Alex/Goi/SF Team           NNUE XV
Crystal 251221      SF Team                Analysis II
ShashChess 21.1     Kiniama/Manzo/SF Team        OTS V   
Stockfish 15        SF Team                   NNUE XIV
Stockfish 15        SF Team                  NNUE XIII
Swordfish 14.8      Dorsz/SF Team             NNUE XII
SF-PB 130322        Alex/SF Team               NNUE XI
Kayra 1.3           Mehmet Karaman/SF Team  Openings I
Cfish 030820        Cfish/SF Team        3600+ Elo III
SF-PB 130322        Alex/SF Team        GUI/Engine III
SF-PB 130322        Alex/SF Team         GUI/Engine II
Cfish 030820        Cfish/SF Team         3600+ Elo II
ef1c0cd86e12.nnue   Lonfom169              EvalFile IV
Blue Marlin 14.4    Dorsz/SF Team           Analysis I
CM11 Innovation     Tom Mente   In Name of Chessmaster
Stockfish 310720    SF Team          MP (3m+1s) Rating
Stockfish 011221    SF Team                     NNUE X
Stockfish 011221    SF Team                    NNUE IX
Kayra 1.0           Mehmet Karaman/SF Team   NNUE VIII
Stockfish 13        SF Team                NON-NNUE IV
Stockfish 13        SF Team               NON-NNUE III
Stockfish 13        SF Team                NON-NNUE II
Cfish 090521        Cfish/SF Team             NNUE VII
a27a526755f0.nnue   Fisherman             EvalFile III
Stockfish 13        SF Team                    NNUE VI
62ef826d1a6d.nnue   SFisGOD                EvalFile II
SugaR AI 1.80       M.Z/K.Omar/SF team          NNUE V
Cfish 030820        Cfish/SF Team          3600+ Elo I
Stockfish 13        SF Team      3000+ (10m+2s) Rating
Stockfish 13        SF Team      Champion of Champions
Stockfish 13        SF Team       3000+ (4m+2s) Rating
Eman 6.91           Khald Omar/SF team         NNUE IV
Eman/SugaR          K.Omar/M.Z/SF team  Handicapped II
Stockfish 080121    SF team                   NNUE III
Stockfish 080121    SF Team                 NON-NNUE I
Stockfish 251220    SF Team          MP (3m+1s) Rating
Stockfish 231120    SF Team          MP (3m+1s) Rating
CF-EXT/CiChess      Cfih/SF team               NNUE II
Stockfish 101120    SF team                   Own NNUE
Lc0 v0262 LS15      JJosh/LCZero team 3000+ (3m+1s) II 
Stockfish XIr4      Mike Byrne/SF Team!     3500+ VIII 
Stockfish 11        SF Team          MP (3m+1s) Rating
S_XPro-NN 1.0       M.Z/SF Team              3500+ VII
Stockfish 11        SF Team       3000+ (3m+1s) Rating
Stockfish 140320    SF Team                   3500+ VI
Stockfish 101219    SF Team          MP (4m+2s) Rating
Stockfish 281119    SF Team         MP (15m+3s) Rating
More Detailş,
1st of all, due to no much free time, I shared some data,
Based on 74 Ratings, organized mainly between 2020-2023
Note that these Ratings are only Engine competitions...
Exc. Duel Book vs Experience III, but all other tours are
Based on stats, which do not include Book Tours or any
Other Testings, such as CPU, GUI, Hash, TC, Strength etc.
Btw, to be honest I'm just shocked (by myself) really e.g
Much BIG data produced..plus all in short period of time!!
And again I wish to say: many thanks tp all Programmers,
Otherwise.. all these valuable rankings would not exist!!
Once more, Congratulations to SCCT Engine Champions!
And as usually, I hope that you like my free activities?)
Sure, they are NOT perfect... but I've done my BEST!

Meanwhile,
Here are some Individual Champion Statistics (since 2020):
Well, current data can be considered as good indicator about
Who are Greatest authors...and here are Best Champ Records:
Stockfish: 34 (Best of Best) / Cfish: 8/ SF-PB: 7/ Kayra: 5
Eman, RapTora and SF Polyglot managed 3 times as Champions!
Actually these Champ stats are since 2020 but if we check over
The past 25 years..then as most frequent Number Ones were
ChessMaster, Fritz, Hiarcs, Shredder, Rybka, Houdini etc.

In same that means / once more we noticed:
-No one can stay on Top forever!
-No one can be on Top by the work of others!
If not so clear:
It's required own original ideas for sure!

And I'm sorry that my postings are not so short,
But due to my tours numbers are not so short!

Ok, that's all for now..take care and stay safe )

Greetings

Re: Sedat Canbaz

Posted: Sat Aug 19, 2023 1:20 pm
by Sedat Canbaz
Image

Hello there )

1st of all,
The above screenshot belongs to SCCT's new live matches:
Eman 8.91 8 Cores vs RapTora 2.2 8 Cores

To be more clear,
I've done my best to adapt to Deeds recent played matches,
E.g Deeds test results are ended +19 Elo in favor for Eman
And very soon, we will see under SCCT conditions...about
What will be the final results: Eman or RapTora will win..?!

On other hand,,
Sure I can't say that SCCT used conditions are same as
Deeds's conditions e.g one of main differences is that
I played via Large Pages as disabled, this is really as
Very important factor! e.g as I stated earlier: in case of
Long-running matches: we can't expect so much accurate
Results (sure I'm referring if as Large Pages as enabled)
Where according to Deeds: he used as Large Pages enabled
And in the current screenshot: NPS values are clear about
The power speed of 2x EPYC 7B12 machine: av. 10.000 kN/s
Sure if we run a test with fast 8 (eight) physical cores!

By the way,
Higher kN/s values (HT/SMT ON) don't mean better in chess!
And for anyone missed (based on various benchmark results),
E.g in chess, physical cores solved better than logical threads:
https://sites.google.com/site/computers ... benchmarks

As another main difference is that (between SCCT and Deeds),
Very soon, I hope to release the latest generated exp files!
Where Deeds is refused to share...not sure why?! but my 6th
Strong sense says that we'll see Deeds test as not so valid!

And please stay tuned, very soon more info, games, exp etc!

Greetings )

Re: Sedat Canbaz

Posted: Sat Aug 19, 2023 1:21 pm
by Sedat Canbaz
Update 2

Well, finely I've managed to share the recent SCCT results...

Note also that,
Both Top engines are played via default settings, exception
Move overhead: 100, Cores 8, Syzygy 4-MEN, 4 GB hash etc.
Plus Eman 8.91 is played via Experience MultiPV as disabled

What I can say more,
35 Elo difference (SCCT vs Deeds) is too much...interesting!
I mean regarding..if we are going to compare with Deeds test:
https://open-chess.org/viewtopic.php?f=4&t=3675

Image

Download all games, exp files, screenshots:
https://mega.nz/file/jkRHGYQZ#nXGR60IR7 ... _Wteol4E6Y


Best,
Sedat

Re: Sedat Canbaz

Posted: Sat Aug 19, 2023 1:22 pm
by Sedat Canbaz
Btw, just small...but as friendly advise,
It's time the new generation, especially who are in new in eng testings
To take some examples (as lessons)...as my current sharing system way!
Because learning process is nothing bad...including humans as well...!!

Greetings

Re: Sedat Canbaz

Posted: Mon Aug 21, 2023 12:02 pm
by Sedat Canbaz
Meanwhile,

As other very important reference about
SCCT's used conditions are faster and
Better in quality than Deeds conditions:

E.g SCCT average move length: 61
In Deeds average move length: 64

If not clear,
E.g as usual, on faster machines,
The overall played length moves
Are ending in less moves...!!
Proved, guaranteed...!!

And that's why I'm much surprised:
2x EPYC 7B12's 8 Cores are faster
Than 2x E5-2660 v3's 40 Threads

And now I wonder much, (in reality)
Deeds testings are with 40 threads?
Is there any PROVE ?) maybe again
Deeds is fooling all of us here ?))

E.g RapTora is used. 32 threads
Where Eman is used: 40 threads

Or other scenario:

E.g RapTora is used. 24 threads
Where Eman is used: 30 threads

Etc. and etc.

Btw, why Deeds is not using Fritz GUIs ?
Frankly I'd prefer Fritz GUI, if nothing else
NPS values are perfectly clear...

On other hand, I used Cutechess GUI
Mainly due to offers a concurrent option...
Otherwise, I'd run Fritz GUI for sure !!

Greetings