TCEC S15, END of an ERA event is much more Brutal than I thought!

[mwyoung]

After 26 games in TCEC SUFI, Leela is leading with 15-11.

My prediction for TCEC was just 55-45 Leela win.
cap4.PNG

Yes, I expected long ago 2:1 Win/Loss ratio for Leela, something on the lines +14 -7 =79. Let's see. I predicted that when the previous S14 was underway with 32930 net.

It IS an end of an era, if anything, the hardware for Leela is below the hardware for Stockfish in the current TCEC. GPU has much more room for improvement than CPU, we will assist in 1-2 years at a complete dominance strength-wise of NN-bases engines. FPGA might come next for chess engines, but it's a bit more involved to make them play chess.

These results are meaningless unless they are 100% Repeatable.....Anyone else been able to recreate LC0's (AlphaZero's) results vs Stockfish engines?

yes....

[Laskos]

After 26 games in TCEC SUFI, Leela is leading with 15-11.

My prediction for TCEC was just 55-45 Leela win.
cap4.PNG

Yes, I expected long ago 2:1 Win/Loss ratio for Leela, something on the lines +14 -7 =79. Let's see. I predicted that when the previous S14 was underway with 32930 net.

It IS an end of an era, if anything, the hardware for Leela is below the hardware for Stockfish in the current TCEC. GPU has much more room for improvement than CPU, we will assist in 1-2 years at a complete dominance strength-wise of NN-bases engines. FPGA might come next for chess engines, but it's a bit more involved to make them play chess.

These results are meaningless unless they are 100% Repeatable.....Anyone else been able to recreate LC0's (AlphaZero's) results vs Stockfish engines?

yes....

I don't even understand what you guys are talking about. Have any of RTX 2060 to 2080ti GPUs, and using a T40 net it will beat Stockfish on 16 core CPU, at least at LTC, from some regular openings. Soon, it will be valid to Blitz and bullet TC (if not already). The only TC Lc0 struggles with is ultra-bullet, where the search tree size is small and would need efficient small batches, but cuDNN is bad with small batches. I guess soon even that will be corrected, as one could use better kernels than the default ones used today, which are more adequate for ultra-fast TC and small tree sizes.

[Raphexon]

I do wonder, how strong is the current net on a CPU.
On CCRL 40/40 Lc0 0.20.1 w36089 is the 35th strongest single core engine. (SF2.2.1 level) On 40/4 it\s significantly weaker.
Wonder how much ELO the latest net (and executable) has in CCRL40/40 conditions.

I guess since Lc0 scales pretty well with nodes she'll be much stronger on modern hardware without an adjusted TC.

[corres]

...
Interesting times, traditional engines will become more like a niche "puzzle solvers".

Yes, for fun of chess engines it is an interesting time - mainly for people who like to watch engine-engine matches.
Peoples who are not only watchers but users of chess engines they will need further the top AB engines like Stockfish, Houdini and Komodo for a long time.

[Laskos]

...
Interesting times, traditional engines will become more like a niche "puzzle solvers".

Yes, for fun of chess engines it is an interesting time - mainly for people who like to watch engine-engine matches.
Peoples who are not only watchers but users of chess engines they will need further the top AB engines like Stockfish, Houdini and Komodo for a long time.

I guess you are the utmost user of chess engines. Are you a high rated ICCF player I am unaware of?
People of all sorts will use many chess tools and software available, I don't think there is a "quality control" on chess engine usage yet.
There are specific tasks about which I have some vague ideas, that's all. I guess for centaur chess, I personally would need as of now Lc0 and Stockfish, but Stockfish mostly as a "proofing tool" for Lc0, a sort of spellchecker, stupid mistakes always occur even when typing something. And maybe for endgames, at least for some time from now.

But I guess you are the best user of chess programs, so please express at large your vast knowledge.

[corres]

...
Interesting times, traditional engines will become more like a niche "puzzle solvers".

...
Yes, for fun of chess engines it is an interesting time - mainly for people who like to watch engine-engine matches.
Peoples who are not only watchers but users of chess engines they will need further the top AB engines like Stockfish, Houdini and Komodo for a long time.

I guess you are the utmost user of chess engines. Are you a high rated ICCF player I am unaware of?
People of all sorts will use many chess tools and software available, I don't think there is a "quality control" on chess engine usage yet.
There are specific tasks about which I have some vague ideas, that's all. I guess for centaur chess, I personally would need as of now Lc0 and Stockfish, but Stockfish mostly as a "proofing tool" for Lc0, a sort of spellchecker, stupid mistakes always occur even when typing something. And maybe for endgames, at least for some time from now.

But I guess you are the best user of chess programs, so please express at large your vast knowledge.

The sarcasm does not suit you well.
I wrote about my opinions and I think it is not only the mine.
If you have any different I will read it with pleasure.

[glennsamuel32]

It's comical to see people rate chess engines by their performance in TCEC.
TCEC is pure entertainment.
Many specialized opening suites already exist to do proper testing.

[Laskos]

It's comical to see people rate chess engines by their performance in TCEC.
TCEC is pure entertainment.
Many specialized opening suites already exist to do proper testing.

Yes, I can get one opening suite, maybe even two. Then all what remains is to play a dual RTX GPU against 43 CPU cores 100 games match at 2 hours + 10s per side, right? Damn, I should have done that long ago, when I will get the opening suite, I will report back the result. Thanks!

[glennsamuel32]

It's comical to see people rate chess engines by their performance in TCEC.
TCEC is pure entertainment.
Many specialized opening suites already exist to do proper testing.

Yes, I can get one opening suite, maybe even two. Then all what remains is to play a dual RTX GPU against 43 CPU cores 100 games match at 2 hours + 10s per side, right? Damn, I should have done that long ago, when I will get the opening suite, I will report back the result. Thanks!

Why should it be 2x RTX gpu's or 44 cores ?
Why should it be 2 hrs per side ? Why not 2m+2sec ?
Why should one do what TCEC does ?
Is TCEC the regulatory body for chess ? Or even Google, that we have to use the 1:1000 nodes ratio ?
In your well detailed tests on this forum, do you test for 2 hours per side, or with 44 cores ?
We don't need the melodrama in your posts.
A proper sample size of games and neutral openings is all that's required.
I would test it but sadly Leela plays crap using a laptop gpu.
When I do get the 2 gpu's, you can be sure I'll test on my own.

[corres]

It's comical to see people rate chess engines by their performance in TCEC.
TCEC is pure entertainment.
Many specialized opening suites already exist to do proper testing.

TCEC has only two issues:
1, !00 games are too few to get an exact Elo
2, They use opening position what were selected on their special demand.
Their viewpoint is differ from the most of engine user.
But there are some important advantage of TCEC opposite to the other test site:
!, Technically well defined and apparent circumstances
2, They use up to date hardware.
3, Peoples can watch continuously the games.