Stockfish Dev Builds FAQ

How often is that site updated?

The site is updated automatically. Every five minutes both repositories are pulled. If there is a new commit on the master branch the binaries are build. This takes some minutes.

What are the differences between all those versions?

  • Windows 32 runs on very old 32 bit versions of windows but it is relativly slow.
  • Windows x64 runs only on 64 bit versions of windows. It is clearly faster.
  • Windows x64 for modern computers additionally requires a cpu which supports the popcnt instruction. Most modern cpus do it. This popcnt instruction speeds up some calculations.
  • Windows x64 for Haswell CPUs additionally uses the bmi2 instruction set which speeds up some calculations further. Until now only Intels Haswell cpus support this instruction set. All future cpus probably will also have that support.
  • Linux … is obvious.

Which version is best for me?

In top-down order. Use the first version that works for you. If you are not sure you can start the program and type “bench”. If it did not crash it works on your computer.

Is there a constant link to the latest version?

Windows x64 for Haswell CPUs
Windows x64 for modern computers
Windows x64
Windows 32
Linux x64 for Haswell CPUs
Linux x64 for modern computers
Linux x64

46 thoughts on “Stockfish Dev Builds FAQ

  1. hi roman
    have you read the comments page lately?
    your decision to remove the popular SSE4.2 compile has created a bit of a stir!

    • SSE4.2 is included with the “modern computers” version. No need for a separate compile. Perhaps the FAQs above should be updated in that regard…

      • JohnnyBob
        Roman’s SSE4.2 compile contains additional instruction groups not present in the “modern” compile, not only SSE4.2 instructions. SSE4.2 is just part of its name

  2. Is anybody working on a AVX2 version?
    This would make a significant improvement possible for Haswell CPU’s

  3. Norton keeps removing your engines from my machine. Can you work with Symantec to make them stop that?

  4. Does somebody know what a 14#### folder is?
    And how to create one?

  5. Is it possible to allow these new versions to use the Stockfish2.3.1.bin openingbook in Arena 3.5? Any solutions welcomed. Thanks

  6. Thanks to your effort, i am constantly running the latest SF since almost half a year. That is great! But since about a week, the syzygy-linux-modern-compile doesnt get provided anymore. Whats up?

  7. Where do I get instructions to install and use the Syzygy Bases version. It looks like there is more work to do than just to download the exe file. Thank you.

  8. What happened to Ronald de Man’s syzygy commit yesterday? It was supposed to merge syzygy options into the standard development version, but not so…

    • Several hours later… the compiled files were changed. The new compilations work OK with syzygy. Thanks!

  9. Why do the last SF builds always create a folder \tmp\stockfishbuilder\src\syzygy\ with a file tbprobe.gcda ?
    In earlier builds (also special syzygy versions) this did not happne, so is this really necessary or can this be removed again?

  10. I started trying some of these development versions of Stockfish, but every one I’ve tried crashes during the game. Any ideas on what’s causing the engine to stop several moves into the game? I’ve enabled the syzygy bases, but I only have the 6 man WDL’s downloaded so far. Are these versions just unstable?

  11. It could be useful if the FAQ could explain exactly what optimisation-related gcc and mingw compiler and linker flags were used to build the binaries or are they just the standard targets in SF’s Makefile (in which case, just stating the target names would be good enough)? It may be that some bench improvements could be found by modifying the flags (or even changing the compiler versions used?).

  12. Comments just went south completely so I can’t respond there. Presumably you’re working on the paging per recent comments…

  13. To stokfish 6 worked with Syzygy is required to download individual files from the tracker? And specify the path to it in the settings of the engine?

  14. Why stockfish is enabled by default only 1 core? Some platforms do not support the function settings of the engine. Is it possible to make engines with automatic on all cores by default? Even experienced users it’s simple – change process priority stockfish on low and the system is not loaded.

  15. I tried using the 20150118 version on Ubuntu 14.04 with syzygy bases, but SF does not seem to recognize the syzygy files, although i defined them under preferences. I tried both Scid and ChessX, but still SF needs to think hard about a completely drawn endgame. No tbhits visible either. Something the matter, or is syzygy not working yet under Linux?

  16. I have discovered wrong versions for download.
    I downloaded the last commit for windows 32 bit from 02.06.2015 12:09.

    Tune pawn shelter/storm
    Bench: 9219343

    But running bench delivers the wrong value 7271152 which is the bench value of the commit directly before this one:

    Tweak backward pawns definition
    Bench: 7271152

    Running this version delives the correct bench 7271152.

    I have compiled the lastest commit and get the correct bench 9219343.

    So its seems the both compiles are the same (but size differs slighlty). Perhaps the problem occurs because the commits have the same filename (there are committed within a few minutes).

    • Mea culpa,

      i checked again from scratch and everything is correct now. I tested this under arena from the command window and something by the installation of my different engines varainats goes wrong.
      Sorry for the false alarm.

  17. hi Everybody,

    anyone can explain what is

    LLR: 2.95 (-2.94,2.94) [0.00,4.00]
    Total: 31679 W: 6183 L: 5912 D: 19584

    LLR: 2.95 (-2.94,2.94) [0.00,4.00]
    Total: 43550 W: 6885 L: 6602 D: 30063

    Bench: 9219343
    Many thanks,


  18. Yes! The AVX2 system works really well on Haswell cpu. That makes significant improvement!

    BMI2 seems like not good enough, Marco.

    To Marco Costaba & developers

    Yes! Would you please add AVX2 system to Haswell cpu? I and many observers recognized this. It works better than bmi2 system. I don’t know why, but that is true.

    Hopefully, developers listen to this advices. You will see how significant improvement it is. Believe me!

    Thanks for share!

  19. To SF Team

    I tested a lot of games on both versions: 15/08/15 & 20/08/15.

    I found out that version 15/08/15 is slightly better than 20/08/15.

    It might be due to aggressive pruning. So we should prepare for a strong one before stage 2 and final stages.

    Thank for your share!

  20. To SF Team

    As a stockfish fan, I tested more than 50 games against komodo 9.1, I perceived that the version 08291516 of Joona kiiski is the strongest one which can knock out komodo 9.1.

    Comparing to version 150815, it is stronger than the version of 150815, and much more stable.

    Could you please replace version 150815 for 08291516 in this tcec season 8?

    I think it is a good decision.

    Good luck to SF team, and the best!

    Thanks for share!

  21. To SF Team,

    Ok! I tested the version 083015. Here is what I realized:

    So far, Only version 08291516 of Joona Kiiski is still the strongest one; it performed much better than the version 083015. I don’t know why?

    Version 083015 is not good enough when competing with komodo 9.1. Only version 08291516 of Joona Kiiski is enough potential against komodo 9.1. The version 083015 for TCEC 8 is not a good choice.

    Here is my personal idea:

    In stage 2 of TCEC 8, I think SF Team should choose a version which is stronger than version 08291516 of Joona.

    If in case of none of any version is better than version 08291516, then only version 08291516 is the best choice.

    Honestly, version 08301520 is very hard to go more further in the later stage.

    Thank for your share!

  22. i think stockfish 13.3.15 is strongest and contempt value=15 is much better than 0 please consider

  23. I was wondering what engine developmental versions are tested against. Are they tested against the latest official release of Stockfish (which would be 6 now), or is each new developmental version tested against the previous developmental version? This is an important question because when I see that the latest developmental version won more games, I’m left wondering…against who?

  24. To the authors of SF

    Due to two lost games recently, I have a comment like this: we should send a suggestion to the TCEC Committee to consider an “exception rule” for all engines, it might be ok.

    Here is the “exception rule”: If two engines meet the minimum score to be considered as a “win”, then the time factor would not be counted. For example, two engines meet the minimum evaluation score is 10 or higher (mate score), then the winning engine is considered as “a win” even though the winning engine is lost by time.

    It does not make any sense that a losing engine gets a win, while the losing engine is so bad. It is not deserved to be claimed a “win”.

    If ALL OF US send comments to the TCEC Committee for repetition, they might reconsider our concerns.

    Ok, this is just my personal idea. Please act immediately!

    SF Fan!


  25. When will be the next version of stockfish released? Already Komodo has released Komodo 9.2 which is stronger than Stockfish 6.

  26. it seems this engine has weaked very much. after 211015 version.

    2 Core: G620 @ 2.60GHz, 4 GB RAM
    Book: Modern SGM 3

    1 Stockfish 211015 64 POPCNT +241 8.0/10
    2 Komodo 8 64-bit -241 2.0/10

    1 Stockfish 231015 64 POPCNT +89 +6/=13/-1 62.50% 12.5/20
    2 Komodo 8 64-bit -89 +1/=13/-6 37.50% 7.5/20

    1 Stockfish 241015 64 POPCNT +27 +5/=18/-3 53.85% 14.0/26
    2 Komodo 8 64-bit -27 +3/=18/-5 46.15% 12.0/26

    1 Stockfish 251015 64 POPCNT +102 +10/=16/-2 64.29% 18.0/28
    2 Komodo 8 64-bit -102 +2/=16/-10 35.71% 10.0/28

    1 Stockfish 261015 64 POPCNT +89 +5/=15/-0 62.50% 12.5/20
    2 Komodo 8 64-bit -89 +0/=15/-5 37.50% 7.5/20

    1 Stockfish 301015 64 POPCNT +89 +7/=11/-2 62.50% 12.5/20
    2 Komodo 8 64-bit -89 +2/=11/-7 37.50% 7.5/20

    1 Stockfish 021115 64 POPCNT +134 +13/=15/-2 68.33% 20.5/30
    2 Komodo 8 64-bit -134 +2/=15/-13 31.67% 9.5/30

    1 Stockfish 031115 64 POPCNT +139 +9/=11/-1 69.05% 14.5/21
    2 Komodo 8 64-bit -139 +1/=11/-9 30.95% 6.5/21

  27. Version stockfish_15120511_x64_bmi2.exe loses on time even when winning.
    Is the lazy_smp problem in this version?

  28. I don’t know why, I’m not a programmer, It’s probably You will be interested, but Stockfish 7 engine loses to SugaR Pro v1.3b POPCNT_64. Which actually is a clone and refined by several people. The game was organized without books and tables, just engine vs engine. Control 1 min in the game, each engine used 6 cores. From 10 games Stockfish 7 lost 3 times. I’m just a fan of your engine and decided that this information will be useful to You

  29. I adjust contempt = 6 is better than 0. It plays better and safer. Please take consideration the contempt value. The high contempt the more risky. So we want somewhat risky, but still secure.

  30. To SF TEAM

    Thank for your share.

    So far, the version 240416 (the former) of Voyageone is the strongest one. It was 880 kb size. This version was removed by the author. I don’t know why…

    After that, it was replaced by a new 240416, but this version was 1.1 mb. The later one, including the latest ones, were slower and weaker than the former one of Voyageone.

    PLEASE investigate the latest ones. Something is not right with the latest codes.

  31. Hi guys I want to get eact Stockfish 200516 but cant find exact this version. can someone help ?

  32. Hi, i read many times now, that the compiles from this site – that i am using and updating on a regular basis – would be slow due to some unintended upgrade of the compiler software. Are you aware of the possibility to downgrade a package and lock it against further upgrades, as described here?
    I would appreciate, if you looked into the option to provide faster compiles once again.

  33. Hello!
    Is it possible to make the following improvements:
    1. To release a version adapted for AMD CPU? The fact is that the compiler from Intel under Windows has a built-in processors and the recognition unit includes a compiled program slowest codes for non-Intel processors. As a result, programs are running on 30% slower than the Intel counterparts.
    2. Is it possible to release a version that supports GPU on video cards (ATI, Nvidia)? And in the future, and the calculator from nVidia Tesla (1 CPU from 1024 to 4096 core)? Can I use both GPU and CPU or just alone?

    Yours sincerely,

Comments are closed.