Nay Lin Tun wrote: ↑Sun Jan 06, 2019 3:46 am
And the newer nets are getting stronger day by day, may be +30 to +50 elo.
I don't believe in this progression.
Elo growth is not deterministic, but for sure not 30-50 Elo points daily
Sorry, bad English.
I did not mean 30 to 50 elo, day by day. (30 to 50 in comparison with the old one playing in TCEC). And new nets are still getting stronger day by day. Yeah in lucky day 2-3 elo at most, but bad nets and good nets are fluctuating.
Test35 (smallnet) has had its final LR drop after 250k steps. It should be interesting to see how it performs at fast TC (blitz/bullet or faster depending on hw) in a day or two. Possibly telling wrt test40 as well.
Nay Lin Tun wrote: ↑Tue Jan 08, 2019 12:53 am
...
This mean Leela become 99.9% best engine within a year!
Beating SF will become 100%.
Especially they will use more powerful hardware for Leela.
But if the clock speed of CPU would increase 3-4 times (as the power of the NIDIA cards)
Stockfish could keep his advantage.
Nay Lin Tun wrote: ↑Tue Jan 08, 2019 12:53 am
...
This mean Leela become 99.9% best engine within a year!
Beating SF will become 100%.
Especially they will use more powerful hardware for Leela.
But if the clock speed of CPU would increase 3-4 times (as the power of the NIDIA cards)
Stockfish could keep his advantage.
I think with the pressure (superiority?) from AMD 2x speed every 2 years is reasonable. GPU-side nvidia essentially has a monopoly on consumer cards and sets the pace of their choice (fortunately not to the extent intel did since the i7 was introduced).
The speedup in nps from 10-20 series was essentially coincidental and won’t be the norm. It’s even possible future non-Ti or even Titan cards will be nerfed if the instructions don’t yield benefits in gaming.
Nay Lin Tun wrote: ↑Tue Jan 08, 2019 12:53 am
Well, this is end of an era for Houdini and Komodo.
This mean Leela become 99.9% best engine within a year!
Beating SF will become 100%.
I do not understand how do you calculate it.
SF may improve in the next year and same for komodo and houdini.
Last edited by Uri Blass on Tue Jan 08, 2019 9:16 am, edited 1 time in total.
Nay Lin Tun wrote: ↑Tue Jan 08, 2019 12:53 am
...
This mean Leela become 99.9% best engine within a year!
Beating SF will become 100%.
Especially they will use more powerful hardware for Leela.
But if the clock speed of CPU would increase 3-4 times (as the power of the NIDIA cards)
Stockfish could keep his advantage.
I think with the pressure (superiority?) from AMD 2x speed every 2 years is reasonable. GPU-side nvidia essentially has a monopoly on consumer cards and sets the pace of their choice (fortunately not to the extent intel did since the i7 was introduced).
The speedup in nps from 10-20 series was essentially coincidental and won’t be the norm. It’s even possible future non-Ti or even Titan cards will be nerfed if the instructions don’t yield benefits in gaming.
The advantage of AMD arises from the cheaper processors.
The growth in frequency is minimal. From the more cores the AB type chess programs get profit only a small.
It was obvious at the time of the announcement of AlphaZero this technique will be the beneficiary of the graphic card higher core numbers. The building of tensor cores into the cards enhances further the benefit.
Jouni wrote: ↑Sun Jan 06, 2019 7:45 pm
Lc0 32425 loses first game in CCC3 stage 3 for Stockfish in endgame with 4x RTX 2080ti. I bet we can safely wait until 2020- before new era!
You cant safely wait until 2020.
Leela is biting back stockfish now!!
Head to Head, equal result now!