Stockfish 30.12. versus SF16 (30.6.)

[Jouni]

6 months progress test. LTC and HERT book:

Code: Select all

Score of SFdev vs stockfish16: 20 - 9 - 471 [0.511]
...      SFdev playing White: 15 - 2 - 233  [0.526] 250
...      SFdev playing Black: 5 - 7 - 238  [0.496] 250
...      White vs Black: 22 - 7 - 471  [0.515] 500
Elo difference: 7.6 +/- 7.3, LOS: 97.9 %, DrawRatio: 94.2 %
500 of 500 games finished.

NCM has +6 Elo after 20000 games and UHO book.

[Ozymandias]

They also use UHO? Stefan's getting much more of a difference out of it. How slow is this LTC?

[Jouni]

60+0.6 th 1 as Stockfish framework.

[Jouni]

Bad timing for test . 31.12. patch

Passed STC:
https://tests.stockfishchess.org/tests/ ... f82b957bb0
LLR: 2.93 (-2.94,2.94) <0.00,2.00>
Total: 12512 W: 3308 L: 3027 D: 6177 Elo +7.80
Ptnml(0-2): 32, 1372, 3189, 1609, 54

Passed LTC:
https://tests.stockfishchess.org/tests/ ... f82b957cd2
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 35946 W: 9128 L: 8809 D: 18009 Elo +3.08
Ptnml(0-2): 19, 3879, 9862, 4190, 23

[Werewolf]

I wish someone would do some work on elo gain from doubling cores and doubling time.

[Modern Times]

I wish someone would do some work on elo gain from doubling cores and doubling time.

Doubling cores from what to what ? Doubling from 2-4 or 4-8 or 8-16 or 32-64 or 64-128 - the answer likely to be different for each.

[Jouni]

Here is https://github.com/official-stockfish/S ... d-elo-gain. But I don't understand anything .

[Werewolf]

I wish someone would do some work on elo gain from doubling cores and doubling time.

Doubling cores from what to what ? Doubling from 2-4 or 4-8 or 8-16 or 32-64 or 64-128 - the answer likely to be different for each.

Yes exactly. And it would be nice to see some modern Stockfish data on this.

The last data I have goes all the way back to Rybka days where speedup was N^0.76

[Werewolf]

Here is https://github.com/official-stockfish/S ... d-elo-gain. But I don't understand anything .

Thank you, I did not know this existed. This is a cup-of-tea, head-in-hands read, but I will take a look.

[Ozymandias]

Bad timing for test . 31.12. patch

Passed STC:
https://tests.stockfishchess.org/tests/ ... f82b957bb0
LLR: 2.93 (-2.94,2.94) <0.00,2.00>
Total: 12512 W: 3308 L: 3027 D: 6177 Elo +7.80
Ptnml(0-2): 32, 1372, 3189, 1609, 54

Passed LTC:
https://tests.stockfishchess.org/tests/ ... f82b957cd2
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 35946 W: 9128 L: 8809 D: 18009 Elo +3.08
Ptnml(0-2): 19, 3879, 9862, 4190, 23

At NCM, the average of the last two devs, including the one above, is exactly 7 Elo above SF 16. That's the total gain for 6 months.

When so widely different approaches are taken to measure Elo, a serious discussion about what it should be is needed.