SF 11 vs T60 Lc0

[Master Om]

This is what happens in chess.com bullet, when newly upgraded Stockfish 11 meet newly upgraded version of T60 lco ( Kiudee setting).

https://ibb.co/GR2Yw1M


Note. T60 become officical version of Lc0 , T40 is a history now.

Which network ?

[AdminX]

This is what happens in chess.com bullet, when newly upgraded Stockfish 11 meet newly upgraded version of T60 lco ( Kiudee setting).

https://ibb.co/GR2Yw1M


Note. T60 become officical version of Lc0 , T40 is a history now.

Which network ?

62078

[Laskos]

Rather than trying to pick a pointless fight, the “new” parameters can be found here: http://talkchess.com/forum3/viewtopic.php?f=2&t=72854

Of course the people that “know”, know that a big grid search for parameters at different time controls is expensive, long winded and not fun. That’s why these “special settings” only come along every once in a while, probably aren’t best or even good at most time controls, and are passed around like an illicit drug or secret handshake.

I do not agree that tuning Lc0 engine parameters is not important. Without fine-tuning, Stockfish would be at least 200 Elo points weaker. The devs should be doing it more intensely and carefully. Kiudee parameters gave me on strong RTX 2070 a boost for SV 256x20 net 1541 of about 40 Elo points, ranging from ultra-fast to long blitz time controls, and having opponents in both SF11 and self-play against the default. It's a serious boost. It is entirely possible that the whole T60 run will surpass T40 run by less than 40 Elo points in Blitz on an RTX GPU, but the computing effort on T60 training run is humongous. Improving the Lc0 engine should be also among priorities. Most of the improvements can come from the Lc0 engine, not from larger and larger nets, at least until some super-GPUs will hit the market.

[Robert Flesher]

This is what happens in chess.com bullet, when newly upgraded Stockfish 11 meet newly upgraded version of T60 lco ( Kiudee setting).

https://ibb.co/GR2Yw1M


Note. T60 become officical version of Lc0 , T40 is a history now.

Which network ?

62078

Hi Ted, why this network? Has it been tested to be strongest? Our was it the latest network at the time?

[AdminX]

This is what happens in chess.com bullet, when newly upgraded Stockfish 11 meet newly upgraded version of T60 lco ( Kiudee setting).

https://ibb.co/GR2Yw1M


Note. T60 become officical version of Lc0 , T40 is a history now.

Which network ?

62078

Hi Ted, why this network? Has it been tested to be strongest? Our was it the latest network at the time?

Hi Robert,

It's the network currently being used in the CCCC on Chess.com

[Nay Lin Tun]

Final= 107 vs 93

https://ibb.co/5607YLJ

[Laskos]

Final= 107 vs 93

https://ibb.co/5607YLJ

+29 -15 for T60
LOS = 98.2%

[Nay Lin Tun]

Final= 107 vs 93

https://ibb.co/5607YLJ

+29 -15 for T60
LOS = 98.2%

Interesting fact.-- Alpha Zero beat Stockfish 8 after 44 million games training( in a controversal setting).

T60 beat SF after 46 million games training in a seemingly fair setting ( they said CPU server cost more because it is a server grade CPU)

T60 is not fully trained yet. T60 can possibly get another +50 elo when she is fully trained.

In my opinion this T60 running on their hardware is probably +50 to +70 elo better than A0 running on 4 TPU in 2017.

[Laskos]

Final= 107 vs 93

https://ibb.co/5607YLJ

+29 -15 for T60
LOS = 98.2%

Interesting fact.-- Alpha Zero beat Stockfish 8 after 44 million games training( in a controversal setting).

T60 beat SF after 46 million games training in a seemingly fair setting ( they said CPU server cost more because it is a server grade CPU)

T60 is not fully trained yet. T60 can possibly get another +50 elo when she is fully trained.

In my opinion this T60 running on their hardware is probably +50 to +70 elo better than A0 running on 4 TPU in 2017.

Yes, I have a similar impression. I think Lc0 T30 got at least the A0 level on the same strong hardware, T40 surpassed it by some 40 Elo points (more with SV nets), and T60 will end up by being about 100 Elo points above A0.

[Nordlandia]

Question for Laskos: is 30m+30s considered fast slow or slow rapid?

I've seen few say that 30+30 is needed to be considered valid.