If you consider the Elo to be a function of the speedup, then clearly the simplest expression relating 1 CPU Elo to 4 CPU Elo is going to be linear, ie, the formula you gave. Note that you can see this also as the first order approximation to a more complicated and unknown function.
This is pretty much what Bob Hyatt's parallel speedup formula expresses as well.
So yeah, if you know how much an engine gains by going from 1 CPU to 4 CPUs, you can calculate its Elo for 4 CPU's from the data for 1 CPU. I don't think that helps you though, because you need the 4 CPU data to calculate the increase in the first place. This increase will not be constant across different engines, due to differences in the parallel algorithm and the way different evaluation terms scale with time control/depth reached.
Elo increasy by scaling factor
Moderators: hgm, Rebel, chrisw
-
- Posts: 2929
- Joined: Sat Jan 22, 2011 12:42 am
- Location: NL