Kai Laskos positional testing

[mbabigian]

Kai, you used a "positional" test suite at one time that gave you a "feel" for how much positional understanding a program had. As I remember, Komodo scored the highest on this test. Have you tried the new NN version of SF to see if it has caught up or surpassed Komodo's score?

I'm curious. Thanks,
Mike.

[Laskos]

Kai, you used a "positional" test suite at one time that gave you a "feel" for how much positional understanding a program had. As I remember, Komodo scored the highest on this test. Have you tried the new NN version of SF to see if it has caught up or surpassed Komodo's score?

I'm curious. Thanks,
Mike.

At 10s/pos
SF NNUE is first, SF classical is the second, Komodo 14 is the third.
Absolute first on RTX GPU is Lc0 with LS15 (by a pretty long shot).

[mbabigian]

Thank you! Very interesting.

What is the numerical spread between the top 4?

[Laskos]

Thank you! Very interesting.

What is the numerical spread between the top 4?

From memory, 10s / position. Lc0 far ahead with 160, then SF NNUE -- 140, SF classical -- 130, Komodo 14 -- 120, Andscacs 0.95 -- 100.
GPU favoring PC, RTX 2070, but only 4 core i7 3.8 GHz.

[Master Om]

Thank you! Very interesting.

What is the numerical spread between the top 4?

From memory, 10s / position. Lc0 far ahead with 160, then SF NNUE -- 140, SF classical -- 130, Komodo 14 -- 120, Andscacs 0.95 -- 100.
GPU favoring PC, RTX 2070, but only 4 core i7 3.8 GHz.

Which net of Lc0 and NNUE SF ?

[Laskos]

Thank you! Very interesting.

What is the numerical spread between the top 4?

From memory, 10s / position. Lc0 far ahead with 160, then SF NNUE -- 140, SF classical -- 130, Komodo 14 -- 120, Andscacs 0.95 -- 100.
GPU favoring PC, RTX 2070, but only 4 core i7 3.8 GHz.

Which net of Lc0 and NNUE SF ?

LS15 for Lc0 and one of the newer SV nets for SF NNUE.