T58 passing LD2

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

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
dkappe
Posts: 272
Joined: Tue Aug 21, 2018 5:52 pm
Full name: Dietrich Kappe

T58 passing LD2

Post by dkappe » Fri Nov 29, 2019 11:18 pm

According to the testers on discord, the T58 run has passed LD2 in strength. My own endgame tests put 58432 in third place behind LE and 42850. Looks like there’s more playing strength to be squeezed out of even the smaller nets.

User avatar
Eduard
Posts: 224
Joined: Fri Oct 26, 2018 10:58 pm
Location: Germany
Full name: Eduard Nemeth
Contact:

Re: T58 passing LD2

Post by Eduard » Sat Nov 30, 2019 2:04 pm

On my GTX 1050 Ti, 1m+2s:
58432 vs LD2 +10 =33 -7

User avatar
Eduard
Posts: 224
Joined: Fri Oct 26, 2018 10:58 pm
Location: Germany
Full name: Eduard Nemeth
Contact:

Re: T58 passing LD2

Post by Eduard » Sat Nov 30, 2019 4:48 pm

Quicktest on 1050 Ti, 1m+2s, GUI Fritz 17, Opening Hert500.pgn by Stefan Pohl:

58427 vs 42850 nach 20 Partien
+2 =12 -6

58447 vs 42850 nach 20 Partien
+6 =11 -3

User avatar
AdminX
Posts: 5182
Joined: Mon Mar 13, 2006 1:34 pm
Location: Acworth, GA
Contact:

Re: T58 passing LD2

Post by AdminX » Mon Dec 02, 2019 7:21 am

In my test results LD2 wins. Using 2x GTX1060, I gave each net their own GPU. Time control was 5'3 No opening books were used.

Code: Select all

Engine Match

Lc0 v0.23 58432 [3] - Lc0 v0.23 LD2 [3]		22.5	-	27.5		(50: +0, =45, -5)		45.00%		Perf = -34 Elo


( 45.0% TP = -34 Elo 68%->[-80,-29] 95%->[-129,-24] 99.7%->[-184,-19] )


50 games played
"Good decisions come from experience, and experience comes from bad decisions."
__________________________________________________________________
Ted Summers

User avatar
Eduard
Posts: 224
Joined: Fri Oct 26, 2018 10:58 pm
Location: Germany
Full name: Eduard Nemeth
Contact:

Re: T58 passing LD2

Post by Eduard » Mon Dec 02, 2019 12:29 pm

Btw, my 50 games are played with Hert500 by Stefan Pohl. It seems that newest T58 nets are even stronger.

User avatar
Eduard
Posts: 224
Joined: Fri Oct 26, 2018 10:58 pm
Location: Germany
Full name: Eduard Nemeth
Contact:

Re: T58 passing LD2

Post by Eduard » Mon Dec 02, 2019 12:40 pm

New strong setting for T58 on slow GPU - - >with LogitQ=true!!

[OPTIONS]
UCI_Chess960=true
Threads=2
MinibatchSize=64
MaxPrefetch=4
LogitQ=true
CPuct=2.700000
FpuValue=0.550000
PolicyTemperature=1.530000
MoveOverheadMs=75000
Slowmover=0.700000

I tested it live on playchess with RTX 1050 Ti. Timecontrol 16+0:


User avatar
AdminX
Posts: 5182
Joined: Mon Mar 13, 2006 1:34 pm
Location: Acworth, GA
Contact:

Re: T58 passing LD2

Post by AdminX » Mon Dec 02, 2019 12:51 pm

Eduard wrote:
Mon Dec 02, 2019 12:29 pm
Btw, my 50 games are played with Hert500 by Stefan Pohl. It seems that newest T58 nets are even stronger.
Hi Eduard,

Yes, I saw that in testing results on Discord. I look forward to tesing with the new setting you posted. :D
"Good decisions come from experience, and experience comes from bad decisions."
__________________________________________________________________
Ted Summers

User avatar
Eduard
Posts: 224
Joined: Fri Oct 26, 2018 10:58 pm
Location: Germany
Full name: Eduard Nemeth
Contact:

Re: T58 passing LD2

Post by Eduard » Mon Dec 02, 2019 1:09 pm

Have fun! But you know, this
MoveOverhead and Slowmover Settings are only for playing on playchess. Not for tests on own Pc.

User avatar
Eduard
Posts: 224
Joined: Fri Oct 26, 2018 10:58 pm
Location: Germany
Full name: Eduard Nemeth
Contact:

Re: T58 passing LD2

Post by Eduard » Mon Dec 02, 2019 4:44 pm

PolicyTemperature=1.520000 seems a little better.

carldaman
Posts: 1736
Joined: Sat Jun 02, 2012 12:13 am

Re: T58 passing LD2

Post by carldaman » Mon Dec 02, 2019 11:11 pm

Thanks Eduard, but on a really slow GPU (AMD Radeon card) it begins to troll as early as the middlegame (!?)
So many suboptimal moves played in a crushing positions...

Post Reply