I think the jump in ELO is very important for human analysis if it is based on improvement in the SMP protocol used. I guess the 20 ELO gain is only for the core engine itself while the SMP improvements are for the multi processor efficiency. Humans analyzing with a computer can gain a tremendous amount of useful time by having an engine using available processors more efficiently. For example a 40 ELO increase due to MP efficiency in a quad core computer (very common today) would be equivalent to a doubling of cores on your hardware for free! You can never have enough speed when you are analyzing ... the more the better. Now if you are using the engine to play against ... to be honest probably a human would have a hard time noticing the difference between SF 4 and SF 6. A humans downfall against a computer will always be due to tactics and unless you can push the engine into a well studied line where almost every move by the computer is predictable, chances are the human will miss some silly tactical oversight and go down quickly after that.
PaulieD wrote:Komodo 1399.48 64-bit 8CPU v SF 180315 64-bit 8CPU (Match)
Stockfish still 9-4 up in the match.
A lot better than TCEC noticeably so....
That is an old version of Komodo.
Komodo are already at 1405.
And 1399.48 was an experimental version with code changes not included in the main development branch. Whether those changes were good or bad or indifferent, whether 1400 would have been any better, no way of knowing just from that 50 games.
Jouni wrote:Isn't +70 ELO for high end computers worth of official compile now!? There has been commercial releases for 10-20 ELO or even less .
To guarantee the quality of a release, we'll release at maximum twice a year.
People who want to have the latest and greatest and can always download and use the latest development snapshot.
And from practical perspective - very few people actually have >= 16 core box in their hands, so the extra release would benefit only very few individuals. And again, they can use the dev snapshot.