SPCC: Testrun of Stockfish 190101 finished

Discussion of computer chess matches and engine tournaments.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
pohl4711
Posts: 928
Joined: Sat Sep 03, 2011 5:25 am
Location: Berlin, Germany
Contact:

SPCC: Testrun of Stockfish 190101 finished

Post by pohl4711 » Thu Jan 10, 2019 11:27 am

Testrun of Stockfish 190101 finished.

Next testrun: Lc0 v0.20.1 N:32463. First testrun of Lc0 for my 3'+1'' mini-ratinglist. Because Lc0 needs the GPU, only 1 game can be played simultaneously. So, it is not possible to play 5000 games (I would need a month for this...) - so Lc0 plays 1200 games (200 games vs. 6 top-engines). Result not before next Thursday. Leela-Ratio is 0.67.

https://www.sp-cc.de


(Perhaps you have to clear your browsercache or reload the website)

lkaufman
Posts: 3251
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: SPCC: Testrun of Stockfish 190101 finished

Post by lkaufman » Thu Jan 10, 2019 5:01 pm

I just noticed the following sentence on your page: "Move Overhead is set to 500ms, if an engine allows to do so."

This would be half of the increment, unless it is a typo (perhaps you meant 50 ms?). We (Komodo) have never used more than 125 ms in any tournament for Overhead, and that was only on a machine using more than 40 threads, where overhead is a concern. I don't know of any GUI which needs more than 50ms on single thread (or even four threads). Using 500 should highly favor those engines that don't have an option for Overhead. So was this a typo, or if not why the huge value for this setting?
Komodo rules!

pohl4711
Posts: 928
Joined: Sat Sep 03, 2011 5:25 am
Location: Berlin, Germany
Contact:

Re: SPCC: Testrun of Stockfish 190101 finished

Post by pohl4711 » Thu Jan 10, 2019 5:24 pm

lkaufman wrote:
Thu Jan 10, 2019 5:01 pm
I just noticed the following sentence on your page: "Move Overhead is set to 500ms, if an engine allows to do so."

This would be half of the increment, unless it is a typo (perhaps you meant 50 ms?). We (Komodo) have never used more than 125 ms in any tournament for Overhead, and that was only on a machine using more than 40 threads, where overhead is a concern. I don't know of any GUI which needs more than 50ms on single thread (or even four threads). Using 500 should highly favor those engines that don't have an option for Overhead. So was this a typo, or if not why the huge value for this setting?
It is no typo. That makes sure, there are no timelosses. If then a timeloss happens, it is clear, that the engine crashed. by the way: the default move overhead of Shredder 13 is 1 (second! = 1000ms). 500ms is not a problem at all. It makes no sense to compare it to the inkrement-value, but to the average game-time. Which is more than 400 seconds = 200 seconds per engine.
500ms move overhead does NOT mean, that the play-speed in the endgame is doubled (I play with 1000ms inkrement). It only means, that the game-thinking-time of the engine is reduced from around 200 seconds to 199.5 seconds, because the engine keeps 500ms on its game-clock. Does not matter at all. And gives definitly no measureable advantage to an engine without 500ms move-overhead.

To clarify what move-overhead means: the GUI gives the clock-value (rest-time) value to engine and starts the engine's thinking. The engine now subs the move-overhead from that given timeor adds it to the already used time: Then the engine believes, it has (in my testings) 500ms less rest-time, than it has in reality. BUT: when the engine plays its move, only the time, the engine really needed for its move is deleted from the game-rest-time in reality (the GUI clock). So, the 500ms move overhead are no loss of time each move played, it is only a “virtual“ time and comes back, when the engine moves again.
Only one move-overhead timevalue is a loss for the whole GUI game-timer, because it has to stay on that clock until game ends. So, in my testings 500ms move overhead means around -0.25% time less for that engine, because the average gametime for an engine in a game of my testruns is around 200“.

pohl4711
Posts: 928
Joined: Sat Sep 03, 2011 5:25 am
Location: Berlin, Germany
Contact:

Re: SPCC: Testrun of Stockfish 190101 finished

Post by pohl4711 » Thu Jan 10, 2019 6:09 pm

To see, how move overhead works, just play some 1'+1“ games in the Fritz or ShredderGUI and give both engines a real high move-overhead value (as high as possible). And then, watch the game live. Then you will see, that the move overhead is not a timeloss for each move, the engines play. It is only a buffer to zero on the overall game-time.
(Otherwise Shredder 13 could not be part of my ratinglist: with its move overhead of 1000ms default and my playtime.of 3'+1“, playing would be impossible for Shredder, when the game gets long, because the move overhead would delete the whole inkrement for each move... but that is not, what happens)

Post Reply