Stockfish 17 Feb 2022 VLTC

[Marek Soszynski]

The 17 Feb 2022 development version of Stockfish was tuned including VLTC (Very Long Time Control). Intrigued as to what that involved in terms of time control and number of games, I looked that up:

Code: Select all

STC  10+0.1  800,000
LTC  60+0.6  30,000
VLTC 180+1.8 800,000

[Eelco de Groot]

Suppose that would be somewhere around 5 minutes per game. Total of 4000000 (4 million) minutes = 66.666,666666666666666666666666667 hours. Say 20 Watt per core, that is 1333 1/3 kWh. It is going to be expensive if you do that in Europe, with current electricity prices.

By coincidence on the spot market the base price is now, today 133 €/MWh. So it costs here in The Netherlands, for professional electricity buyers (industry etc.) 177 €.

[peter]

Suppose that would be somewhere around 5 minutes per game. Total of 4000000 (4 million) minutes = 66.666,666666666666666666666666667 hours. Say 20 Watt per core, that is 1333 1/3 kWh. It is going to be expensive if you do that in Europe, with current electricity prices.

By coincidence on the spot market the base price is now, today 133 €/MWh. So it costs here in The Netherlands, for professional electricity buyers (industry etc.) 177 €.

I read something about "fixed-length matches" regards

[Eelco de Groot]

Hi Peter,

I don't know exactly, or even approximately, how the Fishtest tuning is done. I was just wondering what would be the approximate scale of the electricity involved. Not a big criticism. I mean compared to atom bomb simulations on supercomputers in Russia, North Korea (do they have supercomputers?) or the US, and all the video cards going to Mongolia now, because forbidden in China, for Bitcoin and other cryptocurrency mining -that's why those cards became so expensive- or equally for Blockchaining in general, Fishtest is probably small fish, and rather innocent...

[peter]

Hi Peter,

I don't know exactly, or even approximately, how the Fishtest tuning is done. I was just wondering what would be the approximate scale of the electricity involved. Not a big criticism. I mean compared to atom bomb simulations on supercomputers in Russia, North Korea (do they have supercomputers?) or the US, and all the video cards going to Mongolia now, because forbidden in China, for Bitcoin and other cryptocurrency mining -that's why those cards became so expensive- or equally for Blockchaining in general, Fishtest is probably small fish, and rather innocent...

I'm fully on your side, Eelco, even small fishes can smell

[Paloma]

As you can see on Nextchessmove it drops down.

https://nextchessmove.com/dev-builds

[Jouni]

-8 to SF14.1 . I expect 17.2. patch is soon removed!

[Uri Blass]

-8 to SF14.1 . I expect 17.2. patch is soon removed!

based on what test exactly?

I saw the following

https://tests.stockfishchess.org/tests/ ... 7ec8851462

I believe that
https://nextchessmove.com/dev-builds
is testing only at fast time control and it may be interesting to see comparison of results against old stockfish at longer time control.

[Uri Blass]

Maybe the stockfish framework is going to test is goint to test also against stockfish7 before the patch and after the patch but not 30+0.3 2 cores but
180+1.8 1 core or 60+0.6 8 cores

[Vinvin]

I hope dev test at 10s+0.1s will stop. Too short TC can drive SF in a wrong direction as this test shows.
May be minimum time should be 30s+0.3s.