jp wrote: ↑Sun May 31, 2020 3:03 am
What are the reported nodes per second TCEC gets for Leela on those 4 GPUs and SF on those 176 cores?
You do realize, that the NPS figures for AB and NN programs have nothing to do with each other.
Yes, of course, I realize. (We have been in discussions with others in other threads about this topic before, in case you don't remember.) But for a given NN size, the figures will reveal whether the hardware is "fair" according to different criteria people have suggested. (Not that any of them is my first preference.)
Thanks for the numbers. I'll have to see if I can do the comparison. (I have numbers for SF, not K, and regular NN size.) My guess is the the NN engine is being given far "too much" hardware power (by those criteria), which I guess is not surprising.
I don't think we have nearly enough data to know for sure what is fair with this particular harware setup. What we see here is 44 cores per GPU and it must be fairly close or we would see nothing but blowouts.
Because it would take a thousand games to know what is fair, and nobody has patience for that, I guess we cannot achieve it.
I think Ed's fastlist won't work either because it is calibrated at low depths. But it might be worth a try.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
Dann Corbit wrote: ↑Mon Jun 01, 2020 1:34 am
I don't think we have nearly enough data to know for sure what is fair with this particular hardware setup. What we see here is 44 cores per GPU and it must be fairly close or we would see nothing but blowouts.
Because it would take a thousand games to know what is fair, and nobody has patience for that, I guess we cannot achieve it.
I think Ed's fastlist won't work either because it is calibrated at low depths. But it might be worth a try.
Well, I don't know what you think is a good criterion for "fair". How would you use the hypothetical thousand games to decide what's fair?
All I was talking about is others' ideas of fair: e.g. energy use, price, balanced PC, ...
Those give different "fair" ratios of SF nodes: NN nodes. I found out the numbers in another thread (but not for Komodo, and not for larger NN sizes), because it seemed not much point everyone keeping on talking about those ideas without anyone going ahead and putting in numbers.
Like, to get a fair match against the world champion, they'd need to get drunk, start with 10 times less time on the clock, and be punched by a boxer after every move. Then I might even beat them.
A/B engines and NNs are different architectures, any kind of suggested "fair" metric is going to be arbitrary.
Ovyron wrote: ↑Tue Jun 02, 2020 2:29 pm
Like, to get a fair match against the world champion, they'd need to get drunk, start with 10 times less time on the clock, and be punched by a boxer after every move. Then I might even beat them.
They (the world champion.) From "They" (gender neutral pronoun.)
Basically I used "they" instead of "he" to refer to the world champion to avoid a male pronoun that'd mean a woman couldn't be world champion (so my statement wouldn't make sense as soon as a woman became world champion; I wanted my post to be ageless and genderless but it just led to confusion )
Ovyron wrote: ↑Tue Jun 02, 2020 9:05 pm
They (the world champion.) From "They" (gender neutral pronoun.)
Basically I used "they" instead of "he" to refer to the world champion to avoid a male pronoun that'd mean a woman couldn't be world champion (so my statement wouldn't make sense as soon as a woman became world champion; I wanted my post to be ageless and genderless but it just led to confusion )
Sure woman can become world champion, between women. Between man, hm, I guess the projected age of solar systems is kind of too short for that event to happen .
It's equally probable as for example Chiron winning the current TCEC event.
If Judit Polgar had been born at a different time, it is conceivable that she could have been world champion.
Of course, if a frog had wings, he would not scrape his butt on the rocks.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
Ovyron wrote: ↑Tue Jun 02, 2020 9:05 pm
They (the world champion.)
It still needs clarification. You wrote this.
Ovyron wrote: ↑Tue Jun 02, 2020 2:29 pm
Like, to get a fair match against the world champion, they'd need to get drunk, start with 10 times less time on the clock, and be punched by a boxer after every move. Then I might even beat them.
So you haven't said who is the person or thing getting the fair match against the world champion. Do you mean other humans? That would be completely UNfair. You appear to have mixed up "fair" with "equal outcome". The match needs to be completely unfair to get a roughly equal outcome against a random human. In a fair match, the vastly superior player ends up with a vastly superior score.
jp wrote: ↑Wed Jun 03, 2020 5:36 am
So you haven't said who is the person or thing getting the fair match against the world champion. Do you mean other humans?
I meant me.
Also, I was probably exaggerating, I think my chances would be fair if I got 10 minutes, and the world champion gives a simul to me and 9 other players, and they (the world champion) gets 1 minute in each board. Hopefully the other 9 players are top 10 in the world
Anyway, my point is that from the get-go, maybe I was born with a brain that wasn't as adept at playing chess as the world's champion. That's unfair! Asking that they play in worse conditions to have a fair match against me makes as much sense as asking Stockfish to use a slower CPU or Leela to use a weaker GPU, for "fairness" (but when you achieve fairness you get a 50% outcome from them, which would be unjust for the stronger one )
I predict a woman will be world chess champion in the next 20 years. What I have no idea about is if she has been born yet...
)
