Top Three Engines Are Essentially Equal Strength!

[Sean Evans]

Only ten Elo rating points between the top three programs is quite amazing at this strength level Very exciting times in the computer chess world! We need a round-robin playoff between these three programs, may be 100 games It would potentially make for some great chess game to review and analyze!

CCRL Elo List

Stockfish 8 64-bit 4CPU Elo 3390

Houdini 5.01 64-bit 4CPU Elo 3387

Komodo 10.3 64-bit 4CPU Elo 3380

Cordially,

Sean

[Leo]

I think on powerful hardware like that used in the TCEC the difference is greater.

[Jouni]

But no more if Houdini 6 makes another 200 ELO jump .

[Werewolf]

But no more if Houdini 6 makes another 200 ELO jump .

For that to happen Stockfish has to make a 200 Elo jump first.

[Milos]

I think on powerful hardware like that used in the TCEC the difference is greater.

Lol, sure, the "law of augmented returns".
The longer the TC, the more difference between engines.

Sound like a commercial for mentally challenged. The more money you spent the richer you'd be .

[gerold]

But no more if Houdini 6 makes another 200 ELO jump .

For that to happen Stockfish has to make a 200 Elo jump first.

you got that right. SF leads the way.

[mjlef]

Sean,

In case you missed it, we release Komodo 10.4 in March. It has tested as considerably stronger than Komodo 10.4.

Mark

[Dann Corbit]

I think on powerful hardware like that used in the TCEC the difference is greater.

Lol, sure, the "law of augmented returns".
The longer the TC, the more difference between engines.

Sound like a commercial for mentally challenged. The more money you spent the richer you'd be .

If it is true, it is extremely surprising.

I would expect that under longer and longer time control you would see more and more draws and the Elo values would converge.

If an engine gets larger and larger Elo advantage at bigger time control or with a big thread count that would really be astonishing.

[BrendanJNorman]

Sean,

In case you missed it, we release Komodo 10.4 in March. It has tested as considerably stronger than Komodo 10.4.

Mark

My copy is about equal strength.

[lkaufman]

Sean,

In case you missed it, we release Komodo 10.4 in March. It has tested as considerably stronger than Komodo 10.4.

Mark

My copy is about equal strength.

On the CCRL 40/40 single thread list, Komodo 10.4 is 34 elo ahead of 10.3 after 770 games. It doesn't have enough 4 cpu games yet to talk about the difference there. Overall I think our 20 elo estimated gain was conservative.

Regarding the effect of longer time controls (and/or more cores), if one engine is stronger because of more efficient code then the difference will shrink with more time. But if one scales better than another, either due to better chess knowledge or better pruning/reducing, it is possible for it to gain elo relative to the other engine with more time. It seems pretty clear that both Stockfish and Komodo scale better than Houdini with more time; between SF and Komodo the evidence is mixed.