Of course, it is now time for Honey 12

[peter]

for those that like to over indulge in honey ...

For those of them having old Xeon X5670- CPUs only, a SSE4.1-popcnt compile would be even greater yet maybe.

Sorry, could not resist, thanks for your continuous fine work of art, MIke.

Hinting regards

[MikeB]

for those that like to over indulge in honey ...

For those of them having old Xeon X5670- CPUs only, a SSE4.1-popcnt compile would be even greater yet maybe.

Sorry, could not resist, thanks for your continuous fine work of art, MIke.

Hinting regards

For you Peter, coming right up - it would be just a few minutes.

[MikeB]

for those that like to over indulge in honey ...

For those of them having old Xeon X5670- CPUs only, a SSE4.1-popcnt compile would be even greater yet maybe.

Sorry, could not resist, thanks for your continuous fine work of art, MIke.

Hinting regards

The Modern compile with SSE4.1 and popcnt

https://www.dropbox.com/t/3rRNzXJ3mQN7WjnX -> 7 day expiration

and AMD Ryzen - AVX2 with bmi
https://www.dropbox.com/l/AAAjG0-2ybhbS ... -_yc_nCeN4 - 6 days left

if there is any other requests for a specific build (OS and CPU instruction type) , you can request here and I will see what I can do..the Pi is out of the question — it will take too long — if it would work at all.

Also , these are not compressed — why or why not ? — Ronald's syzygy code has such good compression built in, making them zip files or upx'd gains almost nothing — about 10%.

[peter]

Also , these are not compressed — why or why not ? — Ronald's syzygy code has such good compression built in, making them zip files or upx'd gains almost nothing — about 10%.

Yet I wouldn't have needed them as much as I wouldn't have needed four times the same net. But who cares for 240Mb disk space, if you then have four of the best compiles for Xeon X5670 ever.

A silly question: if the compile loads an other net by UCI- option- path, does it then load the built-in-net into RAM yet too without using it?
I mean, doesn't it have to load the whole engine including a default-embedded-net anyhow?
And does the other one (not embedded) net get into RAM at all too?

Thankful for very fine compiles and support of unwashed-user-questions regards

[MikeB]

Also , these are not compressed — why or why not ? — Ronald's syzygy code has such good compression built in, making them zip files or upx'd gains almost nothing — about 10%.

Yet I wouldn't have needed them as much as I wouldn't have needed four times the same net. But who cares for 240Mb disk space, if you then have four of the best compiles for Xeon X5670 ever.

A silly question: if the compile loads an other net by UCI- option- path, does it then load the built-in-net into RAM yet too without using it?
I mean, doesn't it have to load the whole engine including a default-embedded-net anyhow?
And does the other one (not embedded) net get into RAM at all too?

Thankful for very fine compiles and support of unwashed-user-questions regards

Those are excellent questions and I am not expert enough to answer them. My best conjecture is that some , if not all of the embeded net would be loaded at startup. But as soon as you start using a different net, that net would quickly be cached into memory and into RAM. That is how a well designed OS should work, how well that works I could not tell you.

[MikeB]

This a rough estimate as how these engines may stack up to Stockfish 11r99 - the first 3 are pretty close to each other and BD is 80 Elo behind , which is consistent with where it stood relative to the the engines prior to NNUE, 11r99 signifying the exe built from last commit prior to the Stockfish NNUE commit - it is now roughly 140 Elo gained since that last commit prior to NNUE - of course this is single core/single thread testing - saying rough because the games below were not all at the same TC and only Oki-Maguro played Stockfish-11r99

Code: Select all

# bay
version 0058, Copyright (C) 1997-2016 Remi Coulom and updated by Michael Byrne.
compiled Jul 25 2020 09:18:55.
This program comes with ABSOLUTELY NO WARRANTY.
This is free software, and you are welcome to redistribute it
under the terms and conditions of the GNU General Public License.
See http://www.gnu.org/copyleft/gpl.html for details.

ResultSet>rp c:/cluster.mfb/pgn/09051332.pgn
7500 game(s) loaded
ResultSet>rp c:/cluster.mfb/pgn/09030015.pgn
13500 game(s) loaded
ResultSet>elo
ResultSet-EloRating>mm 0 1
00:00:00,00
ResultSet-EloRating>r
Rank Name                   Rating   Δ     +    -     #     Σ    Σ%     W    L    D   W%    =%   OppR
---------------------------------------------------------------------------------------------------------
   1 Bluefish-12-1739        3146   0.0    7    7  3000 1746.5  58.2  835  342 1823  27.8  60.8  3089
   2 Honey-12-1739           3144   2.2    7    7  3000 1737.0  57.9  801  327 1872  26.7  62.4  3089
   3 Oki-Maguro-12-1739      3138   5.6    4    4  9000 5769.0  64.1 3412  874 4714  37.9  52.4  3039
   4 Black-Diamond-12-1739   3058  79.9    8    8  3000 1259.5  42.0  395  876 1729  13.2  57.6  3110
   5 Stockfish-11r99         3014  44.4    4    4  9000 2988.0  33.2  893 3917 4190   9.9  46.6  3133
---------------------------------------------------------------------------------------------------------
  Δ = delta from the next higher rated opponent
  # = number of games played
  Σ = total score, 1 point for win, 1/2 point for draw

ResultSet-EloRating>los
                       Bl Ho Ok Bl St
Bluefish-12-1705          67 96100100
Honey-12-1705          32    90100100
Oki-Maguro-12-1705      3  9   100100
Black-Diamond-12-1705   0  0  0   100
Stockfish-11r99         0  0  0  0
ResultSet-EloRating>

will run some games now with top3 vs current dev Stockfish ...since none of these engines are a functional equivalent to current dev-Stockfish —
the Honey here all the fat variety - with some ( 4 men and the KR?vKR 5 Men set) EGTB embedded — but all of my games are being adjudicated with 6 men EGTB anyway — so I would expect little impact ... but who knows...

Edit: The net tested here was incorrectly sighted originally as the 1705 net, It is the 1739 net , or also known as "NN-308d71810dff.nnue"
(308d71810dff *20200903-1739.bin)

If you want to know for sure which net you are using run " shasum -a 256 *.bin | cut -b 13-64 --complement" note: Sergio and I use the ".bin", suffix, SF team decided it needed to use ".nnue" as the suffix , which I thought was silly , but they thought it was awesome ... oh well...

[Marek Soszynski]

Exe's are compressed via UPX , so total download for 5 engines is only 62MB. The engines will run in their compressed format with little loss of speed.

More details and exe's here.
https://github.com/MichaelB7/Stockfish/releases/tag/v12

Will be adding more exe's over the next couple of days for a variety of CPUs and OS's.

Great work!

Could we have a Windows x64 SSSE3 non-PC compile please.

[MikeB]

Exe's are compressed via UPX , so total download for 5 engines is only 62MB. The engines will run in their compressed format with little loss of speed.

More details and exe's here.
https://github.com/MichaelB7/Stockfish/releases/tag/v12

Will be adding more exe's over the next couple of days for a variety of CPUs and OS's.

Great work!

Could we have a Windows x64 SSSE3 non-PC compile please.

What do you mean by "non-PC"?

[Marek Soszynski]

Ah, I got confused between ssse3 (which is what I want) and sse3-popcnt (which at the moment I don't).

[MikeB]

Ah, I got confused between ssse3 (which is what I want) and sse3-popcnt (which at the moment I don't).

OK I will do. Looking like an R1 release THIS WEEK, now that it is about +150 Elo over Stockfish 11.99 ( last DEV-SF 11 before NNUE was introduced)