Will a 1660 ti or a 1080 be considered good enough for Leela ? Or does it require something like a 2080 or 2070 super ?
Free Leela analysis online
Sort:
Abhinav0121 wrote:
Will a 1660 ti or a 1080 be considered good enough for Leela ? Or does it require something like a 2080 or 2070 super ?
2060 is the most cost effective for leela!(Approx 8 cores cpu Stockfish strength)
In fact even decent gpu like 1060 is ok! (4 cores cpu strength)
2080Ti is 16 cores cpu stockfish strength.
I just have a 1660Ti on the pc I am currently using. Thank you for showing me the door to Lc0's realm tho !
Jun 17, 2020
0
#5
goldenwriter wrote:
COMPETITOR SITE ALERT
Nah, it's just a basic position analyzer. (But a very good one...)
Jun 17, 2020
0
#7
I'm not really sure about that analysis. For example, I started moving the pieces and went into the main line of the Ruy Lopez. I'd make a White move, and the eval would be like .7, then I'd make the Black move, and the eval would drop to .1 or .2. That doesn't make sense.
Jun 17, 2020
0
#8
Program evaluators make no sense in general. Look at stockfish. If there is a particular numerical assessment, why two moves later it could be so different. That shows the assessment prior was bad or a mixture of the before and after are bad.
If there is a +0.7 to start with, logically with best play on both sides, the number should never fall below this value. I contend it should actually keep increasing.
Jun 17, 2020
0
#11
toomtoom wrote:
Program evaluators make no sense in general. Look at stockfish. If there is a particular numerical assessment, why two moves later it could be so different. That shows the assessment prior was bad or a mixture of the before and after are bad.
If there is a +0.7 to start with, logically with best play on both sides, the number should never fall below this value. I contend it should actually keep increasing.
Horizon effect.
- toomtoom wrote:
Program evaluators make no sense in general. Look at stockfish. If there is a particular numerical assessment, why two moves later it could be so different. That shows the assessment prior was bad or a mixture of the before and after are bad.
If there is a +0.7 to start with, logically with best play on both sides, the number should never fall below this value. I contend it should actually keep increasing.
As chess is extremely drawish game and good hardware for top engines end up > 95% draw rate, + 0.7 mean practically nothing in any phase of the game.
+0.3 or +0.7 will most likely ended as 0.00 if engines keep searching beyond +50 depth with Tablebase assess. ( +0.7 has slightly higher chance of winning than +0.3 though)
Yeah I agree its evaluations are fishy it jumps wildly in evaluation every move.
The eval bar changes because of the horizon effect. Every time you move, the computer can see further into the game. They have algorithms to go deeper on forcing lines, but not on all lines.
How do you get to the analysis board
The domain is no longer safe and is malware, it will attempt to lock out your device
If you dont have good GPU or if you dont know how to install Leela, you can still analyse leela for free here!