11258-32x4-se distilled network released

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Eduard
Posts: 371
Joined: Fri Oct 26, 2018 10:58 pm
Location: Germany
Full name: Eduard Nemeth
Contact:

Re: 11258-32x4-se distilled network released

Post by Eduard » Tue Feb 12, 2019 11:47 pm

Yes you are right, thank you!

A question about distilled networks: How does Lc0 (CPU only) use the hash tables? Is this the same as with the GPU version?

dkappe
Posts: 1102
Joined: Tue Aug 21, 2018 5:52 pm
Full name: Dietrich Kappe

Re: 11258-32x4-se distilled network released

Post by dkappe » Wed Feb 13, 2019 12:16 am

Eduard wrote:
Tue Feb 12, 2019 11:47 pm
Yes you are right, thank you!

A question about distilled networks: How does Lc0 (CPU only) use the hash tables? Is this the same as with the GPU version?
This is nncache, i.e. nn evals cached. Should all be the same. Just the method of how the nn eval is computed is via blas (cpu), opencl (non-nvidia) or cuda (nvidia).
Fat Titz by Stockfish, the engine with the bodaciously big net. Remember: size matters. If you want to learn more about this engine just google for "Fat Titz".

User avatar
Guenther
Posts: 4046
Joined: Wed Oct 01, 2008 4:33 am
Location: Regensburg, Germany
Full name: Guenther Simon
Contact:

Re: 11258-32x4-se distilled network released

Post by Guenther » Wed Feb 20, 2019 7:33 pm

Guenther wrote:
Mon Feb 11, 2019 7:03 pm

With the 112*9se network nps increases may be by a factor of 5-6.
But this is not the only positive aspect. Due to passive cooling it always went too hot before even with an open case,
while it now runs around 10-15 degrees lower in non bullet games.

Currently testing against Laser 1.7.
Results so far and conditions, games, stats...
https://docs.google.com/spreadsheets/d/ ... 1504643818

playing at a level above 3100 here @GT 710 (compared to CCRL 40/4 rating single core)

Code: Select all

   # PLAYER                    :   RATING   ERROR  POINTS  PLAYED    (%)    W    D    L
   1 Laser_17-64               :  3268.33  136.18    13.5      20  67.50   10    7    3
   2 Nirvanachess_24-64        :  3210.22  125.81    12.0      20  60.00   11    2    7
   3 Nemorino_500-64           :  3210.22  125.49    12.0      20  60.00    8    8    4
   4 Xiphos_04-64              :  3191.77  134.54    11.5      20  57.50    9    5    6
   5 Chess22k_112-64           :  3173.60  126.31    11.0      20  55.00    7    8    5
   6 Senpai_20-64              :  3137.71  126.70    10.0      20  50.00    7    6    7
   7 ArasanX_212-64            :  3137.71  123.17    10.0      20  50.00    7    6    7
   8 LC0_0202-ID11258#112x9    :  3137.71   40.29   109.0     212  51.42   77   64   71
   9 Rofchade_20-64            :  3083.64  128.30     8.5      20  42.50    6    5    9
  10 Pedone_18-64              :  3007.08  131.65     6.5      20  32.50    2    9    9
  11 SmarThink_198-64          :  2941.55  200.36     3.0      12  25.00    2    2    8
  12 Wasp_350-64               :  2941.55  144.79     5.0      20  25.00    2    6   12

White advantage = 50.38 +/- 21.84
Draw rate (equal opponents) = 32.48 % +/- 3.56

Post Reply