Latest Stockfish...

[MikeB]

Latest exe's
Stockfish-0270916 - compiled from latest developmental source on official-stockfish/Stockfish github (https://github.com/official-stockfish/Stockfish)
Stockfish-100116-12y - complied from source, site also has Mac exe's for both versions https://github.com/MichaelB7/Stockfish - includes faster play enhancements when mate is found, limit strength by ELO ( includes 40 different skill levels) and 25% increase in king danger score - increases ELO by ~ 3.

Code: Select all

   1 Stockfish 100116-12y 3121     0  5  5  3680 1974.5 53.7 824 555 2301 22.4 62.5 3095 
   2 Stockfish 270916 64  3117   3.3  5  5  3680 1947.5 52.9 813 597 2270 22.1 61.7 3096

[MikeB]

Latest exe's
Stockfish-0270916 - compiled from latest developmental source on official-stockfish/Stockfish github (https://github.com/official-stockfish/Stockfish)
Stockfish-100116-12y - complied from source, site also has Mac exe's for both versions https://github.com/MichaelB7/Stockfish - includes faster play enhancements when mate is found, limit strength by ELO ( includes 40 different skill levels) and 25% increase in king danger score - increases ELO by ~ 3.

Code: Select all

   1 Stockfish 100116-12y 3121     0  5  5  3680 1974.5 53.7 824 555 2301 22.4 62.5 3095 
   2 Stockfish 270916 64  3117   3.3  5  5  3680 1947.5 52.9 813 597 2270 22.1 61.7 3096

just posted a fix to uci elo limit strength function - it was broken (by me) ...source and exe updated

[MikeB]

In case you missed it , the latest fork of Stockfish has been updated, also added a new UCI option "486-DX" , more info can be found here.

http://www.talkchess.com/forum/viewtopi ... 668#688668

In summary , the option , if enabled, would tell Stockfish it only has 1/xxx the time allocated. This was an effort to simulate an older 486-DX machine with current Stockfish. The results might surprise you. A setting of one would tell Stockfish it only has 1/120th of the time allocated, a setting of 20 would inform Stockfish that it inly has 1/2400th the time allocated. Not tested with micro time controls, settings usage is for a time control of 1 minute and 1 second increment or slower. Typically, the setting you would use to simulate a 486-DX on your machine is nps/100,000 , e.g. my machine benches SF at ~2M nps - 2,000,000 /100,000 = 20 . The setting of 20 informs Stockfish to use 1/2400th the time it was allocated for that move.

[AdminX]

In case you missed it , the latest fork of Stockfish has been updated, also added a new UCI option "486-DX" , more info can be found here.

http://www.talkchess.com/forum/viewtopi ... 668#688668

In summary , the option , if enabled, would tell Stockfish it only has 1/xxx the time allocated. This was an effort to simulate an older 486-DX machine with current Stockfish. The results might surprise you. A setting of one would tell Stockfish it only has 1/120th of the time allocated, a setting of 20 would inform Stockfish that it inly has 1/2400th the time allocated. Not tested with micro time controls, settings usage is for a time control of 1 minute and 1 second increment or slower. Typically, the setting you would use to simulate a 486-DX on your machine is nps/100,000 , e.g. my machine benches SF at ~2M nps - 2,000,000 /100,000 = 20 . The setting of 20 informs Stockfish to use 1/2400th the time it was allocated for that move.

That's an interesting idea. Going back in time ...

[MikeB]

In case you missed it , the latest fork of Stockfish has been updated, also added a new UCI option "486-DX" , more info can be found here.

http://www.talkchess.com/forum/viewtopi ... 668#688668

In summary , the option , if enabled, would tell Stockfish it only has 1/xxx the time allocated. This was an effort to simulate an older 486-DX machine with current Stockfish. The results might surprise you. A setting of one would tell Stockfish it only has 1/120th of the time allocated, a setting of 20 would inform Stockfish that it inly has 1/2400th the time allocated. Not tested with micro time controls, settings usage is for a time control of 1 minute and 1 second increment or slower. Typically, the setting you would use to simulate a 486-DX on your machine is nps/100,000 , e.g. my machine benches SF at ~2M nps - 2,000,000 /100,000 = 20 . The setting of 20 informs Stockfish to use 1/2400th the time it was allocated for that move.

That's an interesting idea. Going back in time ...

Just made another update - changes are explained here - also includes a refresh from the official SF source - a speed optimization of about 2%

http://www.talkchess.com/forum/viewtopi ... 805#688805

Source and latest Mac exe here:

https://github.com/MichaelB7/Stockfish

[royb]

In case you missed it , the latest fork of Stockfish has been updated, also added a new UCI option "486-DX" , more info can be found here.

http://www.talkchess.com/forum/viewtopi ... 668#688668

In summary , the option , if enabled, would tell Stockfish it only has 1/xxx the time allocated. This was an effort to simulate an older 486-DX machine with current Stockfish. The results might surprise you. A setting of one would tell Stockfish it only has 1/120th of the time allocated, a setting of 20 would inform Stockfish that it inly has 1/2400th the time allocated. Not tested with micro time controls, settings usage is for a time control of 1 minute and 1 second increment or slower. Typically, the setting you would use to simulate a 486-DX on your machine is nps/100,000 , e.g. my machine benches SF at ~2M nps - 2,000,000 /100,000 = 20 . The setting of 20 informs Stockfish to use 1/2400th the time it was allocated for that move.

That's an interesting idea. Going back in time ...

Just made another update - changes are explained here - also includes a refresh from the official SF source - a speed optimization of about 2%

http://www.talkchess.com/forum/viewtopi ... 805#688805

Source and latest Mac exe here:

https://github.com/MichaelB7/Stockfish

I must be missing something. I compiled this for Linux and installed the resulting engine in SCID-4.6.4. I ran with throttle zero and noted the time to depth to depth 27. The time was 51 secs for my system.

I set the throttle to 1 and re-tested with the same time (51 secs) to get to depth 27.
I set the throttle to 2 and re-tested with the same time (51 secs) to get to depth 27.

If it's running slower, surely the time to depth should be greater?!

What did I miss? I am using infinite analysis mode if that matters.

Thanks for this option (once I figure out how it works!).

[MikeB]

In case you missed it , the latest fork of Stockfish has been updated, also added a new UCI option "486-DX" , more info can be found here.

http://www.talkchess.com/forum/viewtopi ... 668#688668

In summary , the option , if enabled, would tell Stockfish it only has 1/xxx the time allocated. This was an effort to simulate an older 486-DX machine with current Stockfish. The results might surprise you. A setting of one would tell Stockfish it only has 1/120th of the time allocated, a setting of 20 would inform Stockfish that it inly has 1/2400th the time allocated. Not tested with micro time controls, settings usage is for a time control of 1 minute and 1 second increment or slower. Typically, the setting you would use to simulate a 486-DX on your machine is nps/100,000 , e.g. my machine benches SF at ~2M nps - 2,000,000 /100,000 = 20 . The setting of 20 informs Stockfish to use 1/2400th the time it was allocated for that move.

That's an interesting idea. Going back in time ...

Just made another update - changes are explained here - also includes a refresh from the official SF source - a speed optimization of about 2%

http://www.talkchess.com/forum/viewtopi ... 805#688805

Source and latest Mac exe here:

https://github.com/MichaelB7/Stockfish

I must be missing something. I compiled this for Linux and installed the resulting engine in SCID-4.6.4. I ran with throttle zero and noted the time to depth to depth 27. The time was 51 secs for my system.

I set the throttle to 1 and re-tested with the same time (51 secs) to get to depth 27.
I set the throttle to 2 and re-tested with the same time (51 secs) to get to depth 27.

If it's running slower, surely the time to depth should be greater?!

What did I miss? I am using infinite analysis mode if that matters.

Thanks for this option (once I figure out how it works!).

It does not work like that. First no impact in infinite mode. It's a game only setting. Some settings - say 0 and 1 - still make take the same time on one particular move.

this the code:

Code: Select all

int SlowPC          = Options["Throttle"]; 
if (Options["Throttle"])
optimumTime = optimumTime>>SlowPC;

So it's doing a shift right on optimum time during a game it is playing. You will seen immediate difference when you go from 0 to 5 or 5 to 10. The more subtle differences - say going from 0 to 1 or 0 to 2 will not always be visible on every move due to how SF exits a search - it likes to complete the depth it is working on.

I am making an update based on your question.

[MikeB]

another update to both the modified version:

source here

https://github.com/MichaelB7/Stockfish

mac exe here:

https://github.com/MichaelB7/Stockfish/ ... _MacOS.zip

and the official bleeding edge Stockfish

source here:

https://github.com/official-stockfish/Stockfish

and Mac exe here:

https://github.com/MichaelB7/Stockfish/ ... _MacOS.zip

What's in the modified version (Stockfish-101016-12y)?

As in previous modified versions, it includes fast play when mate is found ( faster play than standard Stockfish) and uci limit strength by elo from the menu.
There is a new parameter called Throttle - settings are 0 to 20 - with zero being full strength and 20 is maximum throttle - if you set it to 20 and set the time control to 6 sec with 0.1 inc , SF will only see a few hundred nodes per move , this weakens SF obviously, but differently than the limit strength by elo. Limit strength by Elo has a random factor where SF will deliberately play sub optimal moves, as compared to using throttle, SF will play the best move it sees- but because the search will be shorter and it may not see beyond a very short horizon - on my machine this tested out about 1650 ELO, not really a push over for an average club player - but it will still make the occasional killer move - very much fun to play against.

In summary , at the setting of 6 sec base with 0.1 sec increment with the Throttle setting at 20, it's similar to playing Stockfish on the earliest of PC's , perhaps a 286 or 386 - maybe even slower. As previously mentioned, it often will only see few hundred nodes per move at that setting. And of course, you could use both the Throttle and UCI Limit Strength by Elo setting to really dumb down Stockfish that a child could play against it.

The evaluation is also tweaked , and, in my modest opinion, it may play slightly better than standard SF at longer time controls, not enough testing to confirm yet...

Code: Select all

78 of 12000 games completed...
time control: 12000+200
Date: 10/11/16 : 00:02:32
78 game(s) loaded
Rank Name                      Rating   Δ     +    -     #     Σ    Σ%     W    L    D   W%    =%   OppR 
------------------------------------------------------------------------------------------------------
   1 Stockfish 101016-12y       3142   0.0   34   34    52   30.5  58.7   15    6   31  28.8  59.6  3081 
   2 Stockfish 101016 64 POPC   3117  24.9   34   34    53   28.5  53.8   12    8   33  22.6  62.3  3093 
   3 Komodo 10.1 64-bit         3041  75.4   35   35    51   19.0  37.3    6   19   26  11.8  51.0  3129 
------------------------------------------------------------------------------------------------------

You may follow along on this 12000 game set in realtime here:

https://www.dropbox.com/s/ytirn3d9cyaxu ... a.txt?dl=1

time control is 2 minutes base with two second increment - updates will average about every 3 minutes ( updates upon every 6 games completed) it will run for a few days as I will be tied up working on the quarterly results...

[MikeB]

final results:

Code: Select all

Rank Name                      Rating   Δ     +    -     #     Σ    Σ%     W    L    D   W%    =%   OppR 
------------------------------------------------------------------------------------------------------
   1 Stockfish 101016-12y       3127   0.0    3    3  8000 4464.5  55.8 1995 1066 4939  24.9  61.7  3087 
   2 Stockfish 101016 64 POPC   3123   3.7    3    3  8000 4404.0  55.0 1938 1130 4932  24.2  61.6  3088 
   3 Komodo 10.1 64-bit         3050  73.6    3    3  8000 3131.5  39.1 1037 2774 4189  13.0  52.4  3125 
------------------------------------------------------------------------------------------------------
  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>d
   1 Stockfish 101016-12y      3127 8000.0 (4464.5 : 3535.5)
                                    4000.0 (2010.5 : 1989.5) Stockfish 101016 64 POPC  3123
                                    4000.0 (2454.0 : 1546.0) Komodo 10.1 64-bit        3050
   2 Stockfish 101016 64 POPC  3123 8000.0 (4404.0 : 3596.0)
                                    4000.0 (1989.5 : 2010.5) Stockfish 101016-12y      3127
                                    4000.0 (2414.5 : 1585.5) Komodo 10.1 64-bit        3050
   3 Komodo 10.1 64-bit        3050 8000.0 (3131.5 : 4868.5)
                                    4000.0 (1546.0 : 2454.0) Stockfish 101016-12y      3127
                                    4000.0 (1585.5 : 2414.5) Stockfish 101016 64 POPC  3123
ResultSet-EloRating>los
                          St St Ko
Stockfish 101016-12y         92100
Stockfish 101016 64 POPC   7   100
Komodo 10.1 64-bit         0  0   
ResultSet-EloRating>

[MikeB]

another update to both the modified version:

source here: https://github.com/MichaelB7/Stockfish

mac exe here: https://github.com/MichaelB7/Stockfish/ ... MacOS.zip

and the official bleeding edge Stockfish

source here: https://github.com/official-stockfish/Stockfish

and Mac exe here: https://github.com/MichaelB7/Stockfish/ ... _MacOS.zip

What's in the modified version (Stockfish-102816-14y)?

As in previous modified versions, it includes fast play when mate is found ( faster play than standard Stockfish) and uci limit strength by elo from the menu. It has a parameter called Throttle - settings are 0 to 20 - with zero being full strength and 20 is maximum throttle - if you set it to 20 and set the time control to 6 sec with 0.1 inc , SF will only see a few hundred nodes per move , this weakens SF obviously, but differently than the limit strength by elo. Limit strength by Elo has a random factor where SF will deliberately play sub optimal moves, as compared to using throttle, SF will play the best move it sees- but because the search will be shorter and it may not see beyond a very short horizon - on my machine this tested out about 1650 ELO, not really a push over for an average club player - but it will still make the occasional killer move - very much fun to play against.

In summary , at the setting of 6 sec base with 0.1 sec increment with the Throttle setting at 20, it's similar to playing Stockfish on the earliest of PC's , perhaps a 286 or 386 - maybe even slower. As previously mentioned, it often will only see few hundred nodes per move at that setting. And of course, you could use both the Throttle and UCI Limit Strength by Elo setting to really dumb down Stockfish that a child could play against it.

With code submitted by Ronald De Man, this version of SF will handle MulitiPV with the Syzygy tablebases.
if you leave the Syzygy50MoveRule unchecked, it will display all 4 winning moves in MultiPV
[d]2R5/8/8/8/B3n3/3n4/K2k4/8 w - - 0 1

and the 17 winning moves here
[d]2R5/8/8/8/B3n3/3nk3/8/K7 w - - 0 1

with the Syzygy50MoveRule checked, you will see the 20 drawing moves from a game Karjakin-Shirov, 2007
[d]r7/8/2N5/2N2k1b/3K4/8/8/8 b - - 0 1

at micro tc of 15 sec game with 0.25 increment - its about even with official Stockfish

Code: Select all

396 of 600 games completed...
time control: 1500+25
Date: 10/29/16 : 00:13:41
396 game(s) loaded
Rank Name                      Rating   Δ     +    -     #     Σ    Σ%     W    L    D   W%    =%   OppR 
---------------------------------------------------------------------------------------------------------
   1 Stockfish 102816-14y       3141   0.0   19   19   263  153.5  58.4   84   40  139  31.9  52.9  3079 
   2 Stockfish 281016 64 POPC   3136   5.2   19   19   263  151.0  57.4   85   46  132  32.3  50.2  3081 
   3 Komodo 10.1 64-bit         3023 112.8   19   19   266   91.5  34.4   38  121  107  14.3  40.2  3138 
---------------------------------------------------------------------------------------------------------
396 game(s) loaded

will kick off a test at longer time control with many more later tonight, which you will be able to follow in realtime here:

https://www.dropbox.com/s/efmt7qz1bywbc ... a.txt?dl=1