Quick estimate please

[Werewolf]

Hi Guys,

I'm using asmfish (Nov 4th 2016) - the version based on Stockfish 8 I believe - in my Aquarium IDeA projects.

The latest version is Feb 21st 2017.

Can anyone make a rough estimate of the elo gain between the two versions on a single core at VLTC - 10 minutes per move? I know this is like guessing how many atoms are in a golf ball but...

Has there been much improvement? I've read there's not much difference.


Many thanks.

(the reason I'm asking this question is because it takes 4 hours + to change the versions over on different P.Cs networked for IDeA etc)

[Guenther]

Hi Guys,

I'm using asmfish (Nov 4th 2016) - the version based on Stockfish 8 I believe - in my Aquarium IDeA projects.

The latest version is Feb 21st 2017.

Can anyone make a rough estimate of the elo gain between the two versions on a single core at VLTC - 10 minutes per move? I know this is like guessing how many atoms are in a golf ball but...

Has there been much improvement? I've read there's not much difference.


Many thanks.

(the reason I'm asking this question is because it takes 4 hours + to change the versions over on different P.Cs networked for IDeA etc)

The only available estimations are for extreme fast tc. Otherwise it would take months at least to generate useful data.

Here is an interesting comparison done by Stefan Pohl.
source: http://spcc.beepworld.de/index.htm


This shows a gain of around 15 rating points until some version in early January or late December and after this there seems to be stagnation.
(not very surprising - mostly code cleaning and simplifications in January)

[RJN]

One improvement of note for asmFish, made a few weeks ago, was for Syzygy, where DTZ was not always being probed once a DTZ position was reached.

Not sure how much elo that itself is worth, 5-10 maybe

[jpqy]

Here you can follow the change-log from asmFish:

https://github.com/lantonov/asmFish/wiki/Change-log

JP.

[Werewolf]

This shows a gain of around 15 rating points until some version in early January or late December and after this there seems to be stagnation.
(not very surprising - mostly code cleaning and simplifications in January)

Thanks, that explains a lot. Hopefully the elo will now start to rise