LCZero Accomplishments and Goals Thus Far

[FWCC]

1. To reach the level of A0 (estimated 3300 rating in GTX 1060)

2. To reach the level of latest stockfish on GTX 1060 ( estimated 3550+ rating on 8 cores desktop)

3. To surpass the rating of stockfish on GTX 1060.(getting 3600+)



Achievement milestones

1. Finished 10 millions games,

2. Reached 2800+, super GM level, on 1060 GTX.

3. Gofundme got €5000+ donation for the project.



Future Leela

1. Network will soon be expanded into 15x192 and may further expand into 20x256

( exactly as A0)

2. cuDNN or tensorflow implementation will increase the speed/ elo of Leela on NVIDIA cards (?50% ?100% ?200%), ( too bad for AMD cards though).

3. syzgy Tablebase

4. Auto resign will speed up training up to 30%.

It looks like ID 227 was trained ON NEW NETWORK 15x192! A HUGE jump in ELO just happened:http://lczero.org/ not sure though.

Yes!! Tablebases are coming!

I don't think tablebases should be used in the training of course.

[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.
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.

I don't think anybody misses the fact that given a hardware advantage that will basically make the evaluation free, you can increase your elo to your satisfaction.
If Stockfish had its evaluation FGPA'ed it may be a 4000 elo engine but who cares for that anyway ? The fact is Stockfish has been throwing away evaluation features for the sake of speed throughout the years.

Daniel

[JJJ]

Did Leela just made a 100 elo jump thanks to the bigger net ?

[Vinvin]

Did Leela just made a 100 elo jump thanks to the bigger net ?

I just read about that on Discord.
But the 95 elo jump worth only with the same number of nodes analyzed.
As this neural net is bigger ( now: 15x192, before 10x128) the computation is slower. On my computer, the speed goes down by around 40%.
The means around -40 Elo for this biggernet but +95 Elo for this latest version.
+55 Elo overall.
But all this numbers depend on your configuration.

[Daniel Shawul]

Did Leela just made a 100 elo jump thanks to the bigger net ?

I just read about that on Discord.
But the 97 elo jump worth only with the same number of nodes analyzed.
As this neural net is bigger ( now: 15x192, before 10x128) the computation is slower. On my computer, the speed goes down by around 40%.
The means around -40 Elo for this biggernet but +95 Elo for this latest version.
+55 Elo overall.
But all this numbers depend on your configuration.

This perfectly demonstrates the point I am making.
You can pack in more knowledge with bigger and bigger nets, but if you keep on using the same hardware it may be a wash interms of elo.

[hgm]

It might not be a wash in terms of scaling, though. If the selectivity gets better, slow TC will profit where fast TC does not.

And I don't agree on the hardware advantage. A GPU is not dedicated hardware; many people have one, as part of a (admittely high-end) standard system. Alpha-beta engines are allowed to use it too. That they elect not to, is their own fault.

It would be interesting if we could make our own TPU from a FPGA, though.

[Werewolf]

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.

I don't think anybody misses the fact that given a hardware advantage that will basically make the evaluation free, you can increase your elo to your satisfaction.
If Stockfish had its evaluation FGPA'ed it may be a 4000 elo engine but who cares for that anyway ? The fact is Stockfish has been throwing away evaluation features for the sake of speed throughout the years.

Daniel

I don't think this is the right way to look at it.

Stockfish MUST run on the CPU. LCZero can, and does well to, run on a GPU. The GPU can get fantastically fast with (relatively) small amounts of money - under £1000 here in the UK.

But if I want a really, really fast machine to run Stockfish - which basically means dual Xeon - I'm looking at 10x that price.

So I see it the other way round. LCZero is taking advantage of powerful untapped hardware. That's a breakthrough.

[Daniel Shawul]

It might not be a wash in terms of scaling, though. If the selectivity gets better, slow TC will profit where fast TC does not.

And I don't agree on the hardware advantage. A GPU is not dedicated hardware; many people have one, as part of a (admittely high-end) standard system. Alpha-beta engines are allowed to use it too. That they elect not to, is their own fault.

It would be interesting if we could make our own TPU from a FPGA, though.

I think people with high-end GPU systems here are probably gamers; what most people probably have is integrated intel GPU cards that are often slower than the CPU atleast for lczero.
I concur that the future is in LCZero's favour; gaming cards are very cheap, atleast before the bitcoin craze, compared to the TFLOPs they pack in.

[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.
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.

I don't think anybody misses the fact that given a hardware advantage that will basically make the evaluation free, you can increase your elo to your satisfaction.
If Stockfish had its evaluation FGPA'ed it may be a 4000 elo engine but who cares for that anyway ? The fact is Stockfish has been throwing away evaluation features for the sake of speed throughout the years.

Daniel

I don't think this is the right way to look at it.

Stockfish MUST run on the CPU. LCZero can, and does well to, run on a GPU. The GPU can get fantastically fast with (relatively) small amounts of money - under £1000 here in the UK.

But if I want a really, really fast machine to run Stockfish - which basically means dual Xeon - I'm looking at 10x that price.

So I see it the other way round. LCZero is taking advantage of powerful untapped hardware. That's a breakthrough.

I don't disagree. I am only interested in the fact that AlphaZero's achievement is mostly hardware based.

[FWCC]

I agree with you Daniel with these standard engines I think most here have core monsters and not so much GPU concerned.I now must purchase a good Graphics card.








FWCC
Tal Was Correct