LCZero Accomplishments and Goals Thus Far

[Albert Silver]

Sigh..wake me up when it is 2800 elo running on singe CPU core, which is what every other engine uses in rating lists. As far as I am concerned, it is still a 2100 elo engine there.
I see so many excited people giving a hardware advantage to LCzero, like CCLS does for instance uses a GPU for LCzero and single core CPU for the rest of the engines.

Well, to begin with, I remember when Rybka was the first engine to take advantage of the 64-bit environment when every standard OS was 32-bit. It had a big speed-up, and no one was able to do the same at first. Shredder 64-bit came out a couple of months later but with zero speedup. I don't recall people saying that it needed to run in a 32-bit environment like everyone else to be 'fair'.

The advantage you complain about is just sour grapes in my book. For one thing, if CCLS or whomever offer a GPU, then it is up to the authors to take advantage of it, not for the one who is able to, to learn to dumb down his machine for 'fairness'.

Leela is designed to use a GPU for best performance. it is inherent in its design. If it reaches 100 Elo better than everyone else on my computer because it alone can use the GPU to best advantage, while all others are weaker because they are only able to use the CPU, guess how much I (and everyone who analyzes with engines) will care?

If it is designed solely for the GPU,

I never said anything of the sort. I said it takes advantage of the GPU.

On the other hand, Stockfish can perform 3200+ elo on a mobile processor.

Your point is what exactly?

Lol, don't you even recall what you wrote a second ago... Here

Leela is designed to use a GPU for best performance. it is inherent in its design

My point again, if it can percolate through your thick head, is it is not designed for the GPU only.

Huh. Ad Hominems already? You should be less emotional about it. I will repeat, though clearly you are still misreading what I wrote: I never said anything about using only the GPU. Saying it is designed to use a GPU for best performance, in no shape or form means 'only' the GPU.

[mirek]

Sigh..wake me up when it is 2800 elo running on singe CPU core, which is what every other engine uses in rating lists. As far as I am concerned, it is still a 2100 elo engine there.

And on and on as a broken record, full thread of your nonsensical "impressions" that you have been repeatedly bringing up already over the other threads only to have them debunked again and again - and yet it clearly doesn't stop you. Sorry but at this point I think you must be trolling.

[Daniel Shawul]

Sigh..wake me up when it is 2800 elo running on singe CPU core, which is what every other engine uses in rating lists. As far as I am concerned, it is still a 2100 elo engine there.

And on and on as a broken record, full thread of your nonsensical "impressions" that you have been repeatedly bringing up already over the other threads only to have them debunked again and again - and yet it clearly doesn't stop you. Sorry but at this point I think you must be trolling.

Anybody who dares to question and wont fall in line is trolling for you ?? Here for instance it is
Claimed LcO is a 2800 elo engine when my tests and also CEGT tests it seems says it is 2100, that is one big question that needs to be addressed. Whether you like it or not that is the fact at the moment.

[CMCanavessi]

Sigh..wake me up when it is 2800 elo running on singe CPU core, which is what every other engine uses in rating lists. As far as I am concerned, it is still a 2100 elo engine there.

And on and on as a broken record, full thread of your nonsensical "impressions" that you have been repeatedly bringing up already over the other threads only to have them debunked again and again - and yet it clearly doesn't stop you. Sorry but at this point I think you must be trolling.

Anybody who dares to question and wont fall in line is trolling for you ?? Here for instance it is
Claimed LcO is a 2800 elo engine when my tests and also CEGT tests it seems says it is 2100, that is one big question that needs to be addressed. Whether you like it or not that is the fact at the moment.

That could mean 2 things, that the engine is 2100 and not 2800, or that the testing method is completely flawed...

I could also run stockfish 9 in my old Pentium 2 400, does it mean SF is a 2600 engine?

[Uri Blass]

Sigh..wake me up when it is 2800 elo running on singe CPU core, which is what every other engine uses in rating lists. As far as I am concerned, it is still a 2100 elo engine there.

And on and on as a broken record, full thread of your nonsensical "impressions" that you have been repeatedly bringing up already over the other threads only to have them debunked again and again - and yet it clearly doesn't stop you. Sorry but at this point I think you must be trolling.

Anybody who dares to question and wont fall in line is trolling for you ?? Here for instance it is
Claimed LcO is a 2800 elo engine when my tests and also CEGT tests it seems says it is 2100, that is one big question that needs to be addressed. Whether you like it or not that is the fact at the moment.

CEGT does not say that latest version they tested is 2100


Lczero 217 = 5405.12 1CPU (startposition 87kns)
1 LCZero/217 0.7 1CPU 2197 +126 1010100111111101½½½1 13.5/20
2 EveAnn 1.72 2071 -126 0101011000000010½½½0 6.5/20


Lczero 217 = 5405.12 1CPU (startposition 87kns)
1 LCZero 0.7 1CPU 2504 +381 1111111111½111111½10 18.0/20 !?
2 Popochin 4.0 2123 -381 0000000000½000000½01 2.0/20

Lczero 217 = 5405.12 1CPU (startposition 87kns) (+100 in selfplay = +50 against other opponents)
1 Djinn 1.021 x64 2509 +190 ½½11½1½101110½11111½ 15.0/20
2 LCZero 0.7 1CPU 2319 -190 ½½00½0½010001½00000½ 5.0/20

2197 2504 2319 are the numbers that I see for Lczero217
average is rating 2340

[Joost Buijs]

The TPUs are described in good detail in some paper. This mentions chip surface, and the technology used (i.e. how many nm resolution for the lithography).

From the AlphaZero article it is not clear which TPU they used to play the games.

I think the paper was quite explicit about that. (Or perhaps the AlphaGo Zero paper.)

I didn't read the AlphaGo Zero article, and I don't get why the hardware of AlphaGo Zero and AlphaZero should be the same.

I don't see how Stockfish was crippled. But if you think they are lying, everything stops. TPUs could be as real as phasers, the 10 games could have been generated by having Houdini play them at 1 hour/move, etc. I can just as easily show you 1000 fabricated games as 10.

I think Stockfish was crippled because they deprived it from using an opening-book, let it run on a 32 core machine with 64 threads (hyper-threading), used a fixed time per move and deprived it from using an egtb.

Yes they do. They say they play 'a match', meaning one. Every scientist knows that selecting a result when you are supposed to measure something will be considered plain lying.

Every scientist knows..., knowing is something else than doing. I've been working on a university for 26 years and I saw on many occasions that results were polished, cherrypicked or bad results left out to make the final outcome look more favourable. Not to speak about articles with totally fabricated results which also happens if you read the news.

[David Xu]

Every scientist knows..., knowing is something else than doing. I've been working on a university for 26 years and I saw on many occasions that results were polished, cherrypicked or bad results left out to make the final outcome look more favourable. Not to speak about articles with totally fabricated results which also happens if you read the news.

This is known as "academic fraud", and it's a good way to get yourself fired. If you think Google DeepMind resorted to academic fraud, well, have fun with your conspiracy theorizing.

Meanwhile, the real world continues to move forward.

[David Xu]

Sigh..wake me up when it is 2800 elo running on singe CPU core, which is what every other engine uses in rating lists. As far as I am concerned, it is still a 2100 elo engine there.

And on and on as a broken record, full thread of your nonsensical "impressions" that you have been repeatedly bringing up already over the other threads only to have them debunked again and again - and yet it clearly doesn't stop you. Sorry but at this point I think you must be trolling.

Anybody who dares to question and wont fall in line is trolling for you ?? Here for instance it is
Claimed LcO is a 2800 elo engine when my tests and also CEGT tests it seems says it is 2100, that is one big question that needs to be addressed. Whether you like it or not that is the fact at the moment.

We get it. You don't like Leela or AlphaZero. That doesn't mean you need to spam any Leela-related threads with the same things over and over again. People have already heard your concerns, and if we happen to not agree, then no amount of repeating them will get us to change our minds. Doing that requires you to make an actual argument.

To put it bluntly, Daniel, you sound like someone who feels threatened by AlphaZero, and who isn't interested in giving it a fair shake. This may or may not actually be true, but your repeated dismissals of Leela even as it clearly improves in strength, as well as your absurd claims that AlphaZero is mainly a hardware-based achievement, all give the impression that you don't want it to be true that a new paradigm is in town. And believe me, that's an ugly look to be having.

[Nay Lin Tun]

What computer-chess afficidonados will buy is driven by what engines require. They used to by highly-overclocked dual-socket (or more) systems with the best CPUs they could find, and don't care about the graphics. When NN-based engines will be the dominant technology they will by PCs with modest CPUs and the most expensive GPU card. This could actually be cheaper. In the end the only thing that matters is how much Elo you buy for your buck; there is no law that says Chess must be restricted to CPUs.

Of course this will be a bit hard to swallow for those having invested k$20+ on 32 cores and no graphics card, while a basic $1000 PC with a k$5 GPU would play better Chess.

But any 'hardware avantage' should be measured in $$$, (possibly including the price of electricity);, other ways just don't make any sense.

+1


In 10 years later, if quantum computers are available on consumer market and a program running on them is significantly better than current programs running on CPU or GPU, there is no need to let that program run on CPU or GPU.

There wont be any standard way to compare head to head.
People will decide by their own on spending their money on buying a good Quantum computer for $ 5000 or current computers for $5000.

[Daniel Shawul]

Sigh..wake me up when it is 2800 elo running on singe CPU core, which is what every other engine uses in rating lists. As far as I am concerned, it is still a 2100 elo engine there.

And on and on as a broken record, full thread of your nonsensical "impressions" that you have been repeatedly bringing up already over the other threads only to have them debunked again and again - and yet it clearly doesn't stop you. Sorry but at this point I think you must be trolling.

Anybody who dares to question and wont fall in line is trolling for you ?? Here for instance it is
Claimed LcO is a 2800 elo engine when my tests and also CEGT tests it seems says it is 2100, that is one big question that needs to be addressed. Whether you like it or not that is the fact at the moment.

We get it. You don't like Leela or AlphaZero. That doesn't mean you need to spam any Leela-related threads with the same things over and over again. People have already heard your concerns, and if we happen to not agree, then no amount of repeating them will get us to change our minds. Doing that requires you to make an actual argument.

To put it bluntly, Daniel, you sound like someone who feels threatened by AlphaZero, and who isn't interested in giving it a fair shake. This may or may not actually be true, but your repeated dismissals of Leela even as it clearly improves in strength, as well as your absurd claims that AlphaZero is mainly a hardware-based achievement, all give the impression that you don't want it to be true that a new paradigm is in town. And believe me, that's an ugly look to be having.

You can give whatever meaning you like to my questions..

I am not the one here claiming we have reached 2800+. That is a gross misrepresentation of the reality that it performs like a 2100 elo engine on a single core CPU that many testers uses. I am not even expressing a concern here, like for example I did with the tactical problem of MCTS.

If you are interested with the algorithmic improvements like I am, you would take out the hardware from the equation. There were many bogus claims like bigger nets will solve tactics, MCTS is better than alpha-beta etc... So far as I can tell their approach survives only due to a massive acceleration of a very slow eval.

Take for example Giraffe, which had a neural network eval but it was only 3 layers for good reason. Though it may have given better evaluation than Stockfish, it was still 10x slower and was only able to achieve a 2600 elo engine on single core. That is still pretty impressive, and I am pretty sure if Mathew was given the luxury of not worrying about eval speed like A0, it would have significantly improved its strength by piling on more layers. The fact is LCZero may never go out of the 2100 elo zone on single core ever again because every time you double the NN size you are dropping nps by 2x. There were claims that LCZero has surpassed Giraffee -- that is not true at all. If Mathew was not concerned with performance on single core, he would have piled up layers on layers of NN without giving a hoot about its speed.

So no, hardware is a very important aspect of all these.