TCEC S27 Superfinal: Stockfish vs LC0

[Chris Formula]

https://tcec-chess.com/live.html

[Jouni]

Stockfish plays with 35 Mnps, but crystal kibitzers with 500 Mnps Can't they changed?

[Jouni]

SF +3.67 Win 99.7% draw https://tcec-chess.com/#div=sf&game=8&season=27

[Chris Formula]

Stockfish is landslide favorite based on the poll but both engines are having close battle atm.

I mistyped the number of days for the poll. Wanted 12 days instead of 2.

[chesskobra]

In game 12, I didn't understand why SF made so many moves, almost repeating the position, before making the g4 pawn break? I have seen this type of play in many games in the past. It seems that it waits either until approaching 50 move rule or until low on time before finally making a pawn break. Could someone explain this phenomenon?

Also, in games 10 and 12, both of which led to Q vs R endgame, SF could not win despite having a strong position.

[Chris Formula]

Looks like Stockfish will prevail...

[Graham Banks]

Looks like Stockfish will prevail...

Good!

[Jouni]

For more interesting match SF should play with 10 threads! I expect it still wins.

[Jouni]

Note there is now Altsufi Kibitzer Ponder Bonus. Stockfish has 509 threads and Lc0 8xRTX4090 GPUs .

[Jouni]

More specific info
This bonus will be a replay of the Superfinal
Same book
Same Time Control
Same engine versions
Same configurations (except obviously adjust hardware-related options: threads, minibatch size, hash)
Ponder on
Stockfish hardware: CPU: 2x AMD EPYC 9754. RAM: 24x 32GB DDR5
LCZero hardware: CPU: 2x AMD EPYC 9274F. GPU: 8x NVIDIA RTX 4090. RAM: 24x 32GB DDR5
Administered by Windfishballad.

So that makes an order of magnitude or 6x to 8x more nodes for Leela, 10x to 12x more for SF.
Highest quality chess match ever played!

My quess is, that SF scores better than TCEC machine. Lc0 seems not to benefit much from better speed?