Page 1 of 2
Fat Fritz, SV 3010 and the Nimzo-Indian
Posted: Wed Mar 11, 2020 11:06 pm
by dkappe
After testing t60 and FF in the Najdorf, I decided to step up to the big nets: Sergio’s 384x30 nets trained on t60 games. I had to extend the TC so the big net got enough nodes. A bit past halfway. Let’s see how we go.
3... Bb4 plus the most popular lines 1 ply after that. TC 3 min + 2 sec.
RTX 2070. SF11 gets 8 cpu. Lc0 0.24
The test will run until 400.
Code: Select all
1) FFv471 30 : 236 (+32,=191,-13), 54.0 %
vs. : games ( +, =, -), (%) : Diff, SD, CFS (%)
384x30-3010 : 118 ( 6, 108, 4), 50.8 : +27, 11, 99.4
SF11 : 118 ( 26, 83, 9), 57.2 : +30, 12, 99.5
2) 384x30-3010 3 : 236 (+11,=206,-19), 48.3 %
vs. : games ( +, =, -), (%) : Diff, SD, CFS (%)
FFv471 : 118 ( 4, 108, 6), 49.2 : -27, 11, 0.6
SF11 : 118 ( 7, 98, 13), 47.5 : +3, 12, 60.1
I begin to understand why TCEC use such oddball openings. Who wants to see 93% drawn games? But who wants to play oddball openings over the board?
Re: Fat Fritz, SV 3010 and the Nimzo-Indian
Posted: Wed Mar 11, 2020 11:07 pm
by dkappe
Yes, and all the usual games with colors reversed, before anybody asks.
Re: Fat Fritz, SV 3010 and the Nimzo-Indian
Posted: Wed Mar 11, 2020 11:44 pm
by mmt
Nice tournament and a very good result for FF so far.
Re: Fat Fritz, SV 3010 and the Nimzo-Indian
Posted: Thu Mar 12, 2020 7:25 am
by lkaufman
dkappe wrote: ↑Wed Mar 11, 2020 11:06 pm
After testing t60 and FF in the Najdorf, I decided to step up to the big nets: Sergio’s 384x30 nets trained on t60 games. I had to extend the TC so the big net got enough nodes. A bit past halfway. Let’s see how we go.
3... Bb4 plus the most popular lines 1 ply after that. TC 3 min + 2 sec.
RTX 2070. SF11 gets 8 cpu. Lc0 0.24
The test will run until 400.
Code: Select all
1) FFv471 30 : 236 (+32,=191,-13), 54.0 %
vs. : games ( +, =, -), (%) : Diff, SD, CFS (%)
384x30-3010 : 118 ( 6, 108, 4), 50.8 : +27, 11, 99.4
SF11 : 118 ( 26, 83, 9), 57.2 : +30, 12, 99.5
2) 384x30-3010 3 : 236 (+11,=206,-19), 48.3 %
vs. : games ( +, =, -), (%) : Diff, SD, CFS (%)
FFv471 : 118 ( 4, 108, 6), 49.2 : -27, 11, 0.6
SF11 : 118 ( 7, 98, 13), 47.5 : +3, 12, 60.1
I begin to understand why TCEC use such oddball openings. Who wants to see 93% drawn games? But who wants to play oddball openings over the board?
I think that lots of very strong players would be happy to play oddball openings over the board, if they were guaranteed a second game with the same opening and reversed colors, as in your test.
Re: Fat Fritz, SV 3010 and the Nimzo-Indian
Posted: Thu Mar 12, 2020 8:38 am
by corres
lkaufman wrote: ↑Thu Mar 12, 2020 7:25 am
...
I think that lots of very strong players would be happy to play oddball openings over the board, if they were guaranteed a second game with the same opening and reversed colors, as in your test.
It would give a much more established Elo rating list, I think.
Re: Fat Fritz, SV 3010 and the Nimzo-Indian
Posted: Thu Mar 12, 2020 10:21 am
by Ovyron
lkaufman wrote: ↑Thu Mar 12, 2020 7:25 am
I think that lots of very strong players would be happy to play oddball openings over the board, if they were guaranteed a second game with the same opening and reversed colors, as in your test.
Yes! A strategy would be to get into a difficult position that you have dedicated to master, where you already know how to draw but those lines are very difficult to find, so you ensure a drawn game and very high chances for a decisive game in your favor (at least more chances than in a normal game without reversed colors.)
Re: Fat Fritz, SV 3010 and the Nimzo-Indian
Posted: Thu Mar 12, 2020 5:07 pm
by Nelson Hernandez
WHAT oddball openings???
Re: Fat Fritz, SV 3010 and the Nimzo-Indian
Posted: Thu Mar 12, 2020 10:37 pm
by dkappe
The Nimzo-off is complete (sf11cpu8, ff and sv 3010 on rtx 2070).
Code: Select all
1) FFv471 37 : 400 (+51,=322,-27), 53.0 %
vs. : games ( +, =, -), (%) : Diff, SD, CFS (%)
384x30-3010 : 200 ( 8, 183, 9), 49.8 : +5, 9, 72.6
SF11 : 200 ( 43, 139, 18), 56.2 : +37, 9, 100.0
2) 384x30-3010 32 : 400 (+39,=337,-24), 51.9 %
vs. : games ( +, =, -), (%) : Diff, SD, CFS (%)
FFv471 : 200 ( 9, 183, 8), 50.2 : -5, 9, 27.4
SF11 : 200 ( 30, 154, 16), 53.5 : +32, 9, 100.0
The nets are in a dead heat, but Fat Fritz has the edge over SF11. Definitely a better performance for 3010 than for t60 in the Najdorf.
Re: Fat Fritz, SV 3010 and the Nimzo-Indian
Posted: Thu Mar 12, 2020 10:59 pm
by dkappe
I’ve put up all the decisive games in the Dorf-off on lichess in 15 studies, Dorf-off 1 thru 15.
https://lichess.org/study/UES6Dwmp
Re: Fat Fritz, SV 3010 and the Nimzo-Indian
Posted: Thu Mar 12, 2020 11:13 pm
by dkappe
There were some duplicate games in the nimzo woodpile. Removing those we get:
Code: Select all
1) FFv471 41 : 333 (+51,=258,-24), 54.1 %
vs. : games ( +, =, -), (%) : Diff, SD, CFS (%)
384x30-3010 : 136 ( 8, 119, 9), 49.6 : +8, 10, 79.1
SF11 : 197 ( 43, 139, 15), 57.1 : +41, 10, 100.0
2) 384x30-3010 33 : 336 (+39,=273,-24), 52.2 %
vs. : games ( +, =, -), (%) : Diff, SD, CFS (%)
FFv471 : 136 ( 9, 119, 8), 50.4 : -8, 10, 20.9
SF11 : 200 ( 30, 154, 16), 53.5 : +33, 10, 100.0