Anyone given Stockfish time odds against itself?
Sort:
I want to know how much hardware affects playing strength. Stockfish vs Stockfish would be apples to apples. Bullet chess would get results faster, since several games may bring needed. Time odds just tests through effect of processing power. I varying memory availability would be another useful test. I bet memory matters much more under longer time controls.
Has anyone plotted win ratio vs time control ratio?
I do know that super grand masters beat 2400 rated opponents even in simultaneous games, maybe 20 at a time, though I'm guessing the details.
As for the grand masters, that is knowledge vs time, not so much strength vs time. But it still shows that reducing time to 5-25% does not reduce rating buy 400 points.
Jan 22, 2018
0
#3
Sounds like that would be a good project for you. :)
The traditional rule of thumb for doubling the speed (or thinking time) results in an increase of about 50 to 70 elo points. However, some recent testing indicates that it might be more than that:
http://talkchess.com/forum/viewtopic.php?t=43134
http://talkchess.com/forum/viewtopic.php?t=42553&topic_view=flat&start=40
Thank you for the links. I also wonder how the extra time is used. Are more tree trunks examined, or are two close choices just tested deeper?
Jan 23, 2018
0
#5
Chesserroo2 wrote:
Thank you for the links. I also wonder how the extra time is used. Are more tree trunks examined, or are two close choices just tested deeper?
I don't know, I don't have detailed knowledge of how engines work. I'd guess a little of both; Probably more of looking deeper at candidate moves, but it might also use some time to find new candidate branches.
4-core with 60 minute is favourite against 8-core with 30 minute.
4-core is still ahead if 8-core use permanent brain.
I wonder how much of the diminishing returns from extra speed is from a lack of increase in memory. As for the 4 and 8 cores, coordinating the extra power may have losses. If I had 256 minutes per move, would I play 400-600 points higher? I think maybe. Could I play at 2300 level if I had 256x256=60000=1000hours per move, 2 months of 16 hour days per move? I think maybe.
Jan 23, 2018
0
#8
As an aside, I'm finding that there aren't too many GUIs that can properly run time odds engine tournaments. I'm running one in ChessGUI right now; I'm pretty sure that Winboard can do it too.
Jan 24, 2018
0
#9
It's early in my time odds tournament, only 50 games done so far. Stockfish 8, 64-bit playing against its clone. 128 MB hash, time control is 40 moves in 4 minutes for Stockfish and 40 moves in 2 minutes for the Stockfish clone.
So far, the clone is scoring 21 out of 50, which equals -56 elo.
The board and time is two seperate component in the game. The board situation is usually more important than time. So time odds is not that severe contra material odds.
Jan 26, 2018
0
#11
I'm going to stop my time odds tournament at 200 games, which I think is enough to get a rough idea. The clone scored 80.5 out of 200, which equals -69 elo.
Jan 26, 2018
0
#12
Chesserroo2 wrote:
I want to know how much hardware affects playing strength. Stockfish vs Stockfish would be apples to apples. Bullet chess would get results faster, since several games may bring needed. Time odds just tests through effect of processing power. I varying memory availability would be another useful test. I bet memory matters much more under longer time controls. Has anyone plotted win ratio vs time control ratio? I do know that super grand masters beat 2400 rated opponents even in simultaneous games, maybe 20 at a time, though I'm guessing the details.
I've tried it in the sense that I've played a KBNK position against both lichess level 8 (Stockfish with limited computer resources) and Stockfish on my PC (less limited). The former is crap, the latter pretty reasonable.