Stockfish 8

I am wondering why is SF8 slower on a faster mashine when anylizing some positions? I tested on I7 4770S and I7 5820K.  4770S has for example 7000 kN/s and 5820K has 9000 kN/s but the dept is 37 plies on 4770S and 35 plies on 5820K. 4770S has samsung evo 840 and 5820K has samsung evo 850, the same operating system win 7, the sam GUI Deep Fritz 14.

Thanks for your answer. Never underestimate the unknown. I have registered today and I haven't play a single game yet.  I am not so bad nor over the board play nor in correspondance chess.

Different branching factors or pruning factors? On the faster machine, it's seeing more nodes/sec, so that part of it makes sense.

Maybe I missunderstud the meaning of kilonodes. I thought that more kilonodes would make analyzing faster and the result would be greater depth in less time which is very important in corr chess.

Both kilonodes and depth parameters depend among other things on the position Stockfish is analyzing. For some positions S8 gets to 30 depth while for some other positions S8 gets only to 20. 

However it is not clear how exactly the parameters are calculated.

I don't think you can set this things, it is intrinsically Stockfish thing.

Depth and speed depends on things like:

• Branching factor
• How likely the positions are punced or reduced with things like late-move-reduction
• Multi-threading
• Delay in printing to the standard outputs
• Whether q-search is necessary
• Transposition table
• Tablebase

I tried the same position on dual xeon E5-2670 (20000 kn/s), but the dept was only 36. Also the mashine wasn't faster on some spots when measuring depth, although 20000kn/s.

I thought that any engine on any mashine analyse all these postions (knodes), but faster with greater number of knodes would do that faster and maybe deeper. It seems unlogical that the mashine with more knodes in slower because of "more work" with more knodes.

In my case I'm talking about the same positions using the same engines and different mashines. Some things should be the same, like branching factor. Different are processors and memory (RAM-hash tables). I74770S has only 8gb RAM, and I75820K has 32gb.

If your 5820K has more nodes per second but fewer plies, it could be going faster than the 4770S.  And it would depend how much is storage is used on each.

4770S has 64gb of 232gb free. 5820K has 46gb of 232gb free. I really don't see the connection between more nodes and fewer plies. Maybe I am wrong but I would say opposite, more nodes more plies.

If it is examining more moves from a position it will have more nodes and may not be able to go as many plies.

I thought that every engine examins all possible moves faster or slower.