Engine Elo estimation for Stockfish 6 4CPU

[JWILD]

Hello,
I noticed that the tournament conditions for the 40/40 list have format (Hashtable size= #CPU's x (128 mb or 256 mb). I wonder how much the Elo of a strong engine (I use Stockfish 6 4CPU) would increase as a function of hashtable size. In other words how good would the play/analysis be for Stockfish 6 4CPU (latest version) running on my computer:
a early 2013 macbook pro with the following specs and settings:
2.7 ghz intel core i7
16 gigs of RAM (I run hashtable at its maximum (i.e. 8192mb)
4 threads (1 per core CPU).
0 contempt
max skill
Basically 4 cores at 2048mb per core.

Also, my friend owns a super high powered Mac Pro with (2.7 ghz 12 core and 64 gigs of RAM). We call it the beast. I'm hoping to test some games on there, though Stockfish won't be able to utilize the machine to its fullest (max threads is 8, max hash size is 16gigs). I'll let you guys know how that goes.
- Jordan

[Kirill Kryukov]

In general hash size has very minor effect on engine performance (outside of extremely small sizes).

The gain from hashtable size saturates at some point, depending on time control and hardware speed. Any engine playing blitz game won't benefit much from a 64 GB hashtable, at least not measurably more than from a 256 MB hashtable. Due to CPU's cache the performance might even be better with smaller hashtable.

[JWILD]

OK. So that applies for Blitz games. Does it also apply for infinite analysis feature for analyzing my own games? Is there a rule of thumb? For instance I analyze most non-trivial moves for about 2 minutes and certain critical moves for as much as 10-15 minutes. Perhaps a larger hash is dictated for the 10-15 minute analysis, a medium hash for 2 minute analysis?

[Ray]

For infinite analysis, I will use 2GB or 4GB of hash. On my machine with 16GB or RAM, then yes 8GB is possible but I haven't tried that.

[Kirill Kryukov]

Perhaps a larger hash is dictated for the 10-15 minute analysis, a medium hash for 2 minute analysis?

Yes, but I think the difference is not significant enough to worry about. 512 MB for 2 minutes and 2 GB for 10-15 minutes should be more than enough, IMO. I don't think it makes sense to keep changing hash size (as it's a distraction from actual analysis and one extra variable to keep in mind, which has a very real cost), so I'd just set it to 1 or 2 GB and forget about it. I think the choice of engine and moves to analyze is incomparably more important. Though I don't play advanced chess, competitive advanced chess players might have different view.

[JWILD]

Cool. Thanks for the help, Kirill and Ray.