Latest Update: 01.10.2022

Hello Chess Friends,

As usually, I am pleased to announce that,

A new idea is born, testing Top NNUE chess engines:

With very weak openings and handicapped positions!

If not so clear, one of Black pawns is missing, see picture)

1st of all,

Who said that the Top NNUE engines (e.g 3700+ Elo)

Are so Drawish...?) sure...under these test conditions,

The situation is different...Draw stats are not 98-99% )

As we see, the overall draw - ratio is minus plus: 52%

Well, just I'd like to point out that, to be honest,

I like the current competition..a lot of 'FUN' for sure!

The matches are so exciting...at least not so boring..

Meanwhile (for anyone interested)

All engines played via TUH suite, it's available Download

Note: TUH opening suite is based on 60 opening positions

Note also that many of TUH openings are av. 1.50+, 2.00+

And the winning perc. of many opening lines can be 80%+

Of course, I am referring about under these test conditions )

E.g vs strong opening lines, the weak suites suffer: Results

Btw, BIG thanks to the creators of TCEC and UHO openings

Without you, all these interesting matches would not exist..!

And a few more notes about determining any engine strength,

As I stated earlier, no doubt that best way with strong openings!

Yes, via short well-tuned openings is not required many games,

For example, approx. 1000 games (per player) is quite enough...

And via strong openings, it's not required testing the engines e.g

The same opening as twice, in other words, what I try to explain.

Elo error-margin is much less via strong lines (than weak lines)

Of course, the engines can be tested via weak openings too..

But then highly recommended thousands of games per player!

Plus, we have to use many various opening test suites as well

Even though e.g according to many testings..no any guarantee,

I mean under these conditions (using weak lines) , what will be

The right and accurate standings...Btw, for anyone missed e.g

Some time ago, I ran many Handicapped tourneys, for Results

Where some of the engines suffer, some are doing better etc.

And as other reference, it's not so bad idea to check: Results

Btw, here the Elo difference is less... between Cfish vs SF-PB

But I expect under optimal cond. Elo diff to be less than 100 Elo

Yes.. and it's clear that (via very weak openings) Cfish suffers..!

On other hand, e.g in older Opening Testings: all are played as

Without repeating the 'same' opening as twice and I think that,

Especially for this reason via UHO opening test is appeared so

Much Elo difference (230+ Elo between the first vs last player)

But via Balsa, even not repeating openings: no such difference!

I short, the testings via Balsa: it's much more accurate for sure!

But just in case of test via short strong openings (NNUE 3700+),

Then the draws 98%, exc. Raubfish SC is close to 87% Results

Ok...and let's back to our current engine testings...)

Really impressive...all of the current NNUE engines are super strong!

And...even under these disadvantage conditions...they are as beast !!

Anyhow, SF-PB managed to be Number ONE and here final standings:

Meanwhile, all played games (in PGN) SCCT Handicapped / NNUE

The Winner: SF-PB 130722 - Congrats to Skynet (Aleks), Stockfish Team!

Rank Name Elo + - games score oppo. draws

1 SF-PB 130722 3771 10 10 1800 55% 3738 50%

2 Stockfish 170922 3769 10 10 1800 54% 3738 51%

3 Polyfish 260822 3768 10 10 1800 54% 3738 51%

4 Swordfish 15.3a 3767 10 10 1800 54% 3738 51%

5 Nebula-Vortex 1.0 3765 10 10 1800 54% 3738 51%

6 Kayra 1.6 3765 10 10 1800 54% 3738 52%

7 SF Polyglot 190922 3762 10 10 1800 53% 3738 51%

8 Corchess 170922 3759 10 10 1800 53% 3739 56%

9 BrainLearn 18 3751 10 10 1800 52% 3739 52%

10 ProteusSF-P 040922 3750 10 10 1800 52% 3739 52%

11 ShashChess 24 3739 10 10 1800 50% 3740 52%

12 Honey 14.1.01 3726 10 10 1800 48% 3741 50%

13 Big Ballz 010922 3725 10 10 1800 48% 3741 53%

14 FlamiingTitz 230822 3702 10 10 1800 44% 3742 53%

15 KomodoDragon 3.1 3662 10 10 1800 38% 3745 47%

16 Cfish 130821 3656 10 10 1800 37% 3745 50%

• Calculated by Bayeselo / ProteusSF fixed to 3750 / BlueName = Last Update

More Details:

The NNUE engines are BMI2, participated via theirs default Evalfiles

Any player's move overhead is changed to 100 (against time forfeits)

Exc. Nebula-Fortex is played as default (due to no such move option)

All played via TUH suite / same openings as twice (reversed colors)

The draw percentage. (based on 14400 games) is not so high: 52%

Via Bayeselo, now it's clear about which engines are more Drawish

For examples, Corchess is most Drawish NN engine (3750+): 56%

Where SF-PB 130722 (3750+) is with lowest draw percentage 50%

KomodoDragon lost 2 games on time again e.g as before: Results

All these testings played just out of curiosity and for fun..and enjoy!

As Final Words: Results / Results 2

This posting is going for the 'Criticizers' you can see that 1m+1s

Or even via 1 Core, the Top NNUE engines are doing not so bad!

For more detailed info, don't hesitate to Contact Me ready to help )

Best,

Sedat