TalkChess.com

mwyoung wrote: ↑Wed Oct 14, 2020 1:00 am

Yes, price per performance is clearly won by Stockfish NNUE I agree. On a 4 core system or a smartphone. And If this is all I had, Stockfish is the clearly winner.

The question is if price and performance is not the only consideration, is Stockfish NNUE clearly the best. And that is still up for debate. As you can stack at bunch of RTX 3090 GPU's. For example! And Stockfish as of today has shown itself lacking in good scaling.

The answer is what is best for what you can afford. If you can not get better then a Lc0 Ratio of greater the 1. I would go with Stockfish. This is testing as of today, and could change with updates of both engines.

All I can tell you is Stockfish 12 and above, is equal to a RTX 2080 TI with Lc0 and the best NN on a 2950x CPU playing with Stockfish. Right now today. with almost a 1 to 1 Lc0 ratio. There is very little difference.

Yes, stack a bunch of RTX3090s and then complain that GPU utilization never reaches 30% because a 10 ghz CPU to support Lc0 doesn't exist.

Dann Corbit wrote: ↑Mon Oct 12, 2020 6:19 pm There is new technology just around the corner that is going to breathe new life into the LC0 category of programs.
AMD is creating a topology where the GPUs and CPUs can have syncrhonized access to the same memory sources.
That removes the one huge problem that exists today for LC0.

Yeah and it's also buying Xilinx.
You really like those fairy tale PR announcements, do you?

Dann Corbit wrote: ↑Tue Oct 13, 2020 12:17 am These statements are true for millionaires but not for ordinary people.
Which is to say, it is a lot cheaper to buy a super high end GPU than to buy a super high end CPU machine.
Of course, you may also have to buy a new power supply.
For example, a 2080 super can be bought for $760

and a 1200W power supply for around $40:

so around $800 gets you extremely high compute power for chess. To get equivalent compute power through a regular computer system would be more expensive.

That's absolute BS (you can't buy 2080s for under 800$, 1200W PS for under 160$ and you need at least 250$ CPU to support that 2080S, that's already price of 3960X on which SF-NNUE obliterates Lc0 on RTX3090) and you seems to be clueless about strength on Lc0 and SFNNUE on equal cost hardware.

Milos wrote: ↑Wed Oct 14, 2020 10:50 pm

Dann Corbit wrote: ↑Tue Oct 13, 2020 12:17 am These statements are true for millionaires but not for ordinary people.
Which is to say, it is a lot cheaper to buy a super high end GPU than to buy a super high end CPU machine.
Of course, you may also have to buy a new power supply.
For example, a 2080 super can be bought for $760

and a 1200W power supply for around $40:

so around $800 gets you extremely high compute power for chess. To get equivalent compute power through a regular computer system would be more expensive.

That's absolute BS (you can't buy 2080s for under 800$, 1200W PS for under 160$ and you need at least 250$ CPU to support that 2080S, that's already price of 3960X on which SF-NNUE obliterates Lc0 on RTX3090) and you seems to be clueless about strength on Lc0 and SFNNUE on equal cost hardware.

I bought 2 2080 supers both for less than 800 each./
The supplied link has one for $760.
Most people already have a PC.
The second link is a 1200 watt power supply for $42
A little searching on amazon will easily find such things.

A3960x system will be well over $3000.
This is an inexpensive one:


Try clicking on the supplied links.
I also assume almost everyone already has a PC of some kind. As long as it has a big enough power supply it should be able to drive the card.

Milos wrote: ↑Wed Oct 14, 2020 10:27 pm

Dann Corbit wrote: ↑Mon Oct 12, 2020 6:19 pm There is new technology just around the corner that is going to breathe new life into the LC0 category of programs.
AMD is creating a topology where the GPUs and CPUs can have syncrhonized access to the same memory sources.
That removes the one huge problem that exists today for LC0.

Yeah and it's also buying Xilinx.
You really like those fairy tale PR announcements, do you?

AMD has not disappointed me yet

Dragon by Komodo Chess looks to be somewhere between Leela and Stockfish level, so there are more than two top engines now.

Hi Dann,

it's really boring to discuss again and again about the nummer 1 and clones.
So many strong chess programs are available with more as 2.800 Elo (CEGT).
And all the other programs from 2.200 - 2.800 Elo are nice to have, or the programs available in the past with a great style of chess.

50 available programs, stronger as World Champion Magnus Carlsen!
So why again and again the same discussion.
What for problems have the people search such discussions?

Honestly:
I don't give a shit ... 3200, 3300 or 18500 Elo.
Never a human will have this strength, never a human will understand the moves but we discuss about it.

Not logical!

The playing style is interesting and most important is:
The programmer should go his own way and not copy code for 150 Elo more!

Such a program is much more interesting as clone number 25 with the same level Stockfish have.

Maybe it need a while others will have more understanding for it and other programs with lesser strength goes more in the center of interest. Such things need a while. End of the day, we are ready with all this, we are playing with Chess Challenger 7 from Fidelity.

Best
Frank

Madeleine or whatever is your right name.
Explain me ...

What is your interest on computer chess!
At the moment I can't understand a lot of your messages!

Frank Quisinsky wrote: ↑Mon Nov 16, 2020 11:14 am Madeleine or whatever is your right name.
Explain me ...

What is your interest on computer chess!
At the moment I can't understand a lot of your messages!

I suppose my original interest was finding strong chess engines that had, in additional to strong tactical play, also strong positional play, especially in closed positions as found in French, Benoni, and King's Indian type positions. For the longest time, none of the traditional alpha-beta engines, not even Stockfish, were able to understand and play such positions well.

Even though Leela existed, it required a good GPU to perform anywhere near Stockfish level, which was unsatisfactory for me as I do not have a good GPU on hand. Furthermore, Leela's search meant that it had worse tactical play than traditional alpha-beta engines. It could miss tactics and blunder at times that traditional alpha-beta engines wouldn't.

But with two years of Leela and none of the other engines even coming close to Leela's positional judgement, it became clear to me that handcrafted evaluations were insufficient to implement the type of positional play I desired from the engines. So I was quite excited when NNUE appeared and engines started adopting it, as finally it was possible to have engines that had both the positional skills as Leela has, and the tactical acumen that the traditional alpha-beta engines like Stockfish has.

In this thread specifically I speculated earlier that if other engines such as Komodo adopted NNUE they may gain a strong enough positional judgement to be on the same level as current Leela, breaking the duopoly of Stockfish and Leela at the top, and as it turns out, when the Komodo team came out with Dragon, tests have shown it to be around or above Leela's strength.

Madeleine Birchfield wrote: ↑Mon Nov 16, 2020 11:49 am

Frank Quisinsky wrote: ↑Mon Nov 16, 2020 11:14 am Madeleine or whatever is your right name.
Explain me ...

What is your interest on computer chess!
At the moment I can't understand a lot of your messages!

In this thread specifically I speculated earlier that if other engines such as Komodo adopted NNUE they may gain a strong enough positional judgement to be on the same level as current Leela, breaking the duopoly of Stockfish and Leela at the top, and as it turns out, when the Komodo team came out with Dragon, tests have shown it to be around or above Leela's strength.

Its my estimation that any top tier engine right now, if NNUE is added, will be stronger than Leela on all but the most unbalanced hardware. Komodo is the first to make the leap, but won't be the last. I've already trained networks that beat Ethereal's evaluation by +140 elo. There is a serious lack of chess people who _understand_ machine learning, to a degree further than rerunning the Stockfish code. But those people who do know are rapidly rising through the ranks. Leela team better be looking at new ideas. RTX 3090s may stave off the future, but thats only because venues refuse to update to modern AMD CPUs.