Re: TCEC S15, END of an ERA event is much more Brutal than I thought!
Posted: Wed May 15, 2019 7:35 pm
Computer Chess Club
https://talkchess.com/
I think that price comparison is wrong.Laskos wrote: ↑Wed May 15, 2019 2:03 pmYes, 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.Nay Lin Tun wrote: ↑Wed May 15, 2019 1:04 pm After 26 games in TCEC SUFI, Leela is leading with 15-11.
My prediction for TCEC was just 55-45 Leela win.
cap4.PNG
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.
So, you define "better hardware" as the hardware which allows for a better chess player on that hardware? I guess you will never be wrong with your assessment of hardware as defined by you.Uri Blass wrote: ↑Wed May 15, 2019 8:08 pmI think that price comparison is wrong.Laskos wrote: ↑Wed May 15, 2019 2:03 pmYes, 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.Nay Lin Tun wrote: ↑Wed May 15, 2019 1:04 pm After 26 games in TCEC SUFI, Leela is leading with 15-11.
My prediction for TCEC was just 55-45 Leela win.
cap4.PNG
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.
From my point of view if it is possible to generate a better chess player with lc0's hardware relative to Stockfish's hardware then lc0's hardware is better.
The better hardware is not known in this case because I do not know what is possible to do with the hardwares.Laskos wrote: ↑Wed May 15, 2019 8:18 pmSo, you define "better hardware" as the hardware which allows for a better chess player on that hardware? I guess you will never be wrong with your assessment of hardware as defined by you.Uri Blass wrote: ↑Wed May 15, 2019 8:08 pmI think that price comparison is wrong.Laskos wrote: ↑Wed May 15, 2019 2:03 pmYes, 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.Nay Lin Tun wrote: ↑Wed May 15, 2019 1:04 pm After 26 games in TCEC SUFI, Leela is leading with 15-11.
My prediction for TCEC was just 55-45 Leela win.
cap4.PNG
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.
From my point of view if it is possible to generate a better chess player with lc0's hardware relative to Stockfish's hardware then lc0's hardware is better.
Lazy SMP is already very good on something like TCEC machine. On 43 cores the effective speed-up I guess is in the range of 20-25. With YBW it was maybe 10-12. I doubt another approach will go much above 30-32 with AB engines. It won't change much against Lc0. It's a bit the opposite, I guess Lc0 scaling to 4 and 8 GPUs can be considerably improved.
These results are meaningless unless they are 100% Repeatable.....Anyone else been able to recreate LC0's (AlphaZero's) results vs Stockfish engines?Laskos wrote: ↑Wed May 15, 2019 2:03 pmYes, 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.Nay Lin Tun wrote: ↑Wed May 15, 2019 1:04 pm After 26 games in TCEC SUFI, Leela is leading with 15-11.
My prediction for TCEC was just 55-45 Leela win.
cap4.PNG
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.
Nay Lin Tun wrote: ↑Wed May 15, 2019 1:04 pm After 26 games in TCEC SUFI, Leela is leading with 15-11.
My prediction for TCEC was just 55-45 Leela win.
cap4.PNG