/Stormphrax

UCI chess engine, with NNUE trained from zero knowledge

Primary LanguageC++GNU General Public License v3.0GPL-3.0

Stormphrax

License
GitHub release (latest by date) Commits since latest release

LGBTQ+ friendly trans rights

a work-in-progress UCI chess and chess960 engine, with NNUE evaluation trained from zero knowledge starting with random weights

this project is a continuation of my HCE engine Polaris

Strength

Version CCRL 40/15 CCRL Blitz CCRL 40/2 FRC CEGT 40/4 CEGT 40/20 MCERL
5.0.0 3506 3623 3866 3501 3461 -
4.1.0 3490 3587 3808 - 3432 -
4.0.0 3477 3567 3778 3440 3419 3542
3.0.0 3409 3492 3693 - 3354 3495
2.0.0 3399 3483 3673 3339 - 3482
1.0.0 3319 3376 3521 3235 - 3346

Features

  • standard PVS with quiescence search and iterative deepening
    • aspiration windows
    • futility pruning
    • history
      • capture history
      • 1-ply continuation history (countermove history)
      • 2-ply continuation history (follow-up history)
      • 4-ply continuation history
    • history pruning
    • internal iterative reduction
    • killers (2 per ply)
    • late move reductions
    • late move pruning
    • mate distance pruning
    • multicut
    • nullmove pruning
    • reverse futility pruning
    • probcut
    • SEE move ordering and pruning
    • singular extensions
      • double extensions
      • triple extensions
      • various negative extensions
    • Syzygy tablebase support
  • NNUE
    • (768x4->1536)x2->1x8 architecture, horizontally mirrored
    • trained from zero knowledge with reinforcement learning from a randomly-initialised network
  • BMI2 attacks in the bmi2 build, otherwise fancy black magic
    • pext/pdep for rooks
    • pext for bishops
  • lazy SMP
  • static contempt

To-do

  • make it stronger uwu

UCI options

Name Type Default value Valid values Description
Hash integer 64 [1, 131072] Memory allocated to the transposition table (in MB).
Clear Hash button N/A N/A Clears the transposition table.
Threads integer 1 [1, 2048] Number of threads used to search.
UCI_Chess960 check false false, true Whether Stormphrax plays Chess960 instead of standard chess.
UCI_ShowWDL check true false, true Whether Stormphrax displays predicted win/draw/loss probabilities in UCI output.
ShowCurrMove check false false, true Whether Stormphrax starts printing the move currently being searched after a short delay.
Move Overhead integer 10 [0, 50000] Amount of time Stormphrax assumes to be lost to overhead when making a move (in ms).
EnableWeirdTCs check false false, true Whether unusual time controls (movestogo != 0, or increment = 0) are enabled. Enabling this option means you recognise that Stormphrax is neither designed for nor tested with these TCs, and is likely to perform worse than under X+Y.
SyzygyPath string <empty> any path, or <empty> Location of Syzygy tablebases to probe during search.
SyzygyProbeDepth spin 1 [1, 255] Minimum depth to probe Syzygy tablebases at.
SyzygyProbeLimit spin 7 [0, 7] Maximum number of pieces on the board to probe Syzygy tablebases with.
EvalFile string <internal> any path, or <internal> NNUE file to use for evaluation.

Builds

vnni512: requires BMI2, AVX-512 and VNNI (Zen 4/Cascade Lake-SP/Rocket Lake and up)
avx512: requires BMI2 and AVX-512 (Skylake-X, Cannon Lake)
avx2-bmi2: requires BMI2 and AVX2 and assumes fast pext and pdep (i.e. no Bulldozer, Piledriver, Steamroller, Excavator, Zen 1, Zen+ or Zen 2)
avx2: requires BMI and AVX2 - primarily useful for pre-Zen 3 AMD CPUs back to Excavator
sse41-popcnt: needs SSE 4.1 and popcnt - for older x64 CPUs

If in doubt, compare the avx2-bmi2 and avx2 binaries and pick the one that's faster. BMI2 will always be faster on Intel CPUs.

Alternatively, build the makefile target native for a binary tuned for your specific CPU (see below)

Note:

  • If you have an AMD Zen 1 (Ryzen x 1xxx), Zen+ (Ryzen x 2xxx) or Zen 2 (Ryzen x 3xxx) CPU, use the avx2 build even though your CPU supports BMI2. These CPUs implement the BMI2 instructions pext and pdep in microcode, which makes them unusably slow for Stormphrax's purposes.

Building

Requires Make and a competent C++20 compiler that supports LTO. GCC is not currently supported, so the usual compiler is Clang. MSVC explicitly does not work.

> make <BUILD> CXX=<COMPILER>
  • replace <COMPILER> with your preferred compiler - for example, clang++ or icpx
    • if not specified, the compiler defaults to clang++
  • replace <BUILD> with the binary you wish to build - native/vnni512/avx512/avx2-bmi2/avx2/sse41-popcnt
    • if not specified, the default build is native
  • if you wish, you can have Stormphrax include the current git commit hash in its UCI version string - pass COMMIT_HASH=on

By default, the makefile builds binaries with profile-guided optimisation (PGO). To disable this, pass PGO=off. When using Clang with PGO enabled, llvm-profdata must be in your PATH.

Credit

Stormphrax makes use of the following libraries:

  • Fathom for tablebase probing, licensed under the MIT license
  • a slightly modified version of incbin for embedding neural network files, under the Unlicense
  • Zstandard for decompressing NNUE files, under GPLv2 (see COPYING)

Stormphrax is tested on this OpenBench instance - thanks to all the people there, SP would be much weaker without your support :3

In no particular order, these engines have been notable sources of ideas or inspiration:

Stormphrax's current networks are trained with bullet. Previous networks were trained with Marlinflow.

The name "Stormphrax" is a reference to the excellent Edge Chronicles :)