November Knockout

[Graham Banks]

(8.0-4.0) Crafty 25.2 64-bit v Quazar 0.4 64-bit

[Graham Banks]

(5.5-6.5) Demolito 2017-08026 64-bit v Rodent III 0.210 64-bit

[Graham Banks]

(5.5-6.5) Deep Junior 13.3 64-bit v Spark 1.0 64-bit

[Graham Banks]

(6.0-6.0) Dirty 30Apr2017 64-bit v Amoeba 2.6 64-bit (GUI decides which engine goes through)

[lucasart]

256mb hash each where possible
40 moves in 25 minutes repeating (adapted for the CCRL)

I think I finally understand why Demolito underperforms dramatically in CCRL 40/40 compared to 40/4. It's probably not a problem of search or eval scalability, but simply of hash pressure. Having the most basic hash replacement scheme is very bad in CCRL 40/40 condition, because the hash pressure there is enormous (ie. 256mb is very small for such a long time control). In turn magnifies the effect of hash replacement, disadvantaging Demolito (whose hash replacement sucks).

I've already improved my hash replacement, and it has shown significant gain at testing conditions that aim to reproduce the CCRL 40/40 hash pressure. Hopefully the next version won't get beaten by Rodent anymore...

[Graham Banks]

(5.5-6.5) Deuterium 14.3.34.130 64-bit v Wasp 2.01 64-bit

[Graham Banks]

(8.0-4.0) Cheng 4.39 64-bit v EXchess 7.97b 64-bit

[Graham Banks]

(4.0-8.0) Thinker 5.4d Inert 64-bit v Spike 1.4 Leiden

[Graham Banks]

(8.0-4.0) Bobcat 8.0 64-bit v Gaviota 1.0 64-bit

[Graham Banks]

QUARTER-FINAL MATCHES

Crafty 25.2 64-bit v Rodent III 0.210 64-bit
Cheng 4.39 64-bit v Amoeba 2.6 64-bit
Spark 1.0 64-bit v Spike 1.4 Leiden
Bobcat 8.0 v Wasp 2.01 64-bit