WCCC 2016 Live games

[Henk]

20 times faster means about 300 ELO points ? So estimated ELO of Jonny there is 2963 + 300 = 3263 ? That would mean it is 30 ELO points stronger than Komodo.

[mehmet karaman]

20 times faster means about 300 ELO points ? So estimated ELO of Jonny there is 2963 + 300 = 3263 ? That would mean it is 30 ELO points stronger than Komodo.

According to my calculations;

42 (elo gain for doubling speed // 10^8 - 10^9 kn/s)
Log(2)20 =4.32

20 x kn/s =42 x 4.32= 181 elo

[cdani]

20 x kn/s =42 x 4.32= 181 elo

Lack of or bad knowledge can limit this more, as the engine even with 2400 cores will not visit many lines that thinks are not promising.

[Laskos]

20 times faster means about 300 ELO points ? So estimated ELO of Jonny there is 2963 + 300 = 3263 ? That would mean it is 30 ELO points stronger than Komodo.

According to my calculations;

42 (elo gain for doubling speed // 10^8 - 10^9 kn/s)
Log(2)20 =4.32

20 x kn/s =42 x 4.32= 181 elo

Close, might be even less. I will re-post what I posted several months ago in another thread. We see that at this level the doubling is probably 30-40 ELO points.


First doubling: 10s+0.1s vs 5s+0.05s -- 148 ELO points
Second doubling: 20s + 0.2s vs 10s+0.1s -- 128 ELO points
Third doubling: 40s+0.4s vs 20s+0.2s -- 110 ELO points

Other data:
Close to eighth doubling -- CEGT 40/20' -- 65 ELO points
Close to ninth doubling -- CCRL 40/40' -- 55 ELO points

Diminishing returns are very visible here. I fitted these results with the relevant curve. Here is the plot of ELO gain per doubling in time and speed:


The red dots are not data, they are predictions, first for Larry's 45'+15'' on 24 cores level against humans in odds matches, for which Larry got as Komodo performing at 3250 FIDE ELO level, second red dot shows TCEC level of 150'+30'' on 24 faster cores. One sees that at those high levels the gain per doubling is below 40 ELO points.

[JJJ]

20 times faster means about 300 ELO points ? So estimated ELO of Jonny there is 2963 + 300 = 3263 ? That would mean it is 30 ELO points stronger than Komodo.

According to my calculations;

42 (elo gain for doubling speed // 10^8 - 10^9 kn/s)
Log(2)20 =4.32

20 x kn/s =42 x 4.32= 181 elo

Close, might be even less. I will re-post what I posted several months ago in another thread. We see that at this level the doubling is probably 30-40 ELO points.


First doubling: 10s+0.1s vs 5s+0.05s -- 148 ELO points
Second doubling: 20s + 0.2s vs 10s+0.1s -- 128 ELO points
Third doubling: 40s+0.4s vs 20s+0.2s -- 110 ELO points

Other data:
Close to eighth doubling -- CEGT 40/20' -- 65 ELO points
Close to ninth doubling -- CCRL 40/40' -- 55 ELO points

Diminishing returns are very visible here. I fitted these results with the relevant curve. Here is the plot of ELO gain per doubling in time and speed:


The red dots are not data, they are predictions, first for Larry's 45'+15'' on 24 cores level against humans in odds matches, for which Larry got as Komodo performing at 3250 FIDE ELO level, second red dot shows TCEC level of 150'+30'' on 24 faster cores. One sees that at those high levels the gain per doubling is below 40 ELO points.

If the final is played with 44 core and Komodo play it, what will be the estimated rating of Kodomo then ? And how much elo time control give to Komodo also ?

[Laskos]

If the final is played with 44 core and Komodo play it, what will be the estimated rating of Kodomo then ? And how much elo time control give to Komodo also ?

I can give a rough FIDE ELO level for the current Komodo at TCEC. An estimate from the data would be 2.5 doublings above FIDE ELO 3250 of Larry Kaufman handicap matches, meaning 80 computer ELO above 3250, or 50 human FIDE ELO above, so close to 3250+50=3300 FIDE ELO for current Komodo TCEC.

Time control doubling now at TCEC for Komodo would mean 35 computer ELO points, hardware doubling 24->48 cores probably only 15 computer ELO points or so (the scaling deteriorates with the number of cores).

[mjlef]

These calculations are based on using all the cores for a search. Jonny uses a lot of the cores to preponder various moves. When it hits a lot, it saves a lot of time. Johanes says when it hits a lot then it wins a lot. If it does not hit as much, it does worse.

I certainly agree there must be a limit in how much extra cores could help. But that is just based on the kinds of things I see other programs and Komodo do now. Cluster control for a chess program is a very interesting topic (hint, anyone have a cluster I can borrow???)

Another issue is we do not know the current Jonny's rating. Looking at TCEC results, it seems to have gained a lot over the last year. Johannes Zwanzger is doing some other very clever things too.

In any case, it is doing pretty well. hats off to Johannes.

Mark

[lkaufman]

First pair of playoff games for first between Komodo and Jonny drawn. Next come faster games.