Skip to content

Stormphrax 8.0.0

Latest

Choose a tag to compare

@Ciekce Ciekce released this 27 Jun 14:06

a whole year!

Stormphrax 8.0.0, now with threat inputs and @JonathanHallstrom's pawn-pawn input technologies, and a lot of other grinding. New network (undertown), multipv fixes, and better performance when running multiple instances at once.

Huge thanks to @87flowers, @KazApps and the aforementioned @JonathanHallstrom for their contributions, along with everyone involved with the threat inputs push (see the PR), and everyone who has contributed compute to our OpenBench instance. This release would not have been remotely possible without huge amounts of compute incredibly generously donated by @KazApps.

Results against Stormphrax 7.0.0

STC:

Elo   | 62.44 +- 4.24 (95%)
Conf  | 8.0+0.08s Threads=1 Hash=16MB
Games | N: 8110 W: 2911 L: 1469 D: 3730
Penta | [46, 461, 1680, 1741, 127]

https://chess.swehosting.se/test/14248/

LTC:

Elo   | 81.93 +- 5.07 (95%)
Conf  | 60.0+0.60s Threads=1 Hash=128MB
Games | N: 4008 W: 1495 L: 567 D: 1946
Penta | [2, 130, 821, 1040, 11]

https://chess.swehosting.se/test/14249/

SMP VLTC (8 threads):

Elo   | 94.51 +- 6.69 (95%)
Conf  | 60.0+0.60s Threads=8 Hash=512MB
Games | N: 2000 W: 775 L: 244 D: 981
Penta | [0, 39, 392, 568, 1]

https://chess.swehosting.se/test/14250/

DFRC LTC:

Elo   | 94.07 +- 7.39 (95%)
Conf  | 60.0+0.60s Threads=1 Hash=128MB
Games | N: 2024 W: 734 L: 199 D: 1091
Penta | [0, 55, 392, 540, 25]

https://chess.swehosting.se/test/14251/

Elo testing done with Fishtest's UHO_Lichess_4852_v1.epd book for standard chess and DFRC_4852_v1.epd for DFRC.
Results are obtained from self-play and will likely not scale perfectly to games against other opponents :)

Where is the avx2 binary?

Almost everyone who downloaded avx2 actually wants avx2-bmi2. If you have an older AMD CPU (Ryzen 3xxx or earlier), use the zen2 binary.

The avx512 binary requires VNNI and VBMI2, and is comparable in its support to Stockfish's avx512icl build. Prefer this one if your CPU supports it, it is significantly faster than avx2-bmi2.

apple-m1 binary provided courtesy of @yl25946.

I strongly recommend using official builds from below or building from source yourself, rather than using third party builds which may have had unknown changes made.

Full Changelog: v7.0.0...v8.0.0