kk wrote:Hi Herbert,
Your example very much illustrates one of the points I had experienced.
I am aware that for every ply that is completed something like an exponential increase in time is required to complete the next ply....however, i can leave the processing overnight and see no update in reporting of analysis where the score for the side to move is high.
....
Yes and aggravating is that the number of nodes does not change (still 5871 kN/s), although TaskM shows 50% load for Komodo.
It hangs in an endless loop?
Power wastage.
Paloma wrote:Here komodo has a similar behavior.
depth 31 in 1:00
depth 32 in 2:40
...
depth 35 in 8:06
...
depth 38 in 28:13
depth 39 in 40:12 and still standing on this depth for 10 hours.
I have noticed another quirk in longer analysis sessions in that the nodes per second does tend to reduce. This happens to both Komodo and Houdini and I wonder whether it may be tied up with hash reaching 100%. I don't always clear hash between analysis of different positions, but I do find by closing and restarting the program this gives a boost to the nps. Not quite sure why this happens but is prevalent with both Komodo and to a lesseer extent Houdini.
I'm under the impression that in endgames the nodes per second is much higher and this is due to fewer pieces on the board but also perhaps to fewer program instructions at this stage of the game for each node reached? Don't know if I am right on this but those are my thoughts to date.
kk wrote:I have noticed another quirk in longer analysis sessions in that the nodes per second does tend to reduce. This happens to both Komodo and Houdini and I wonder whether it may be tied up with hash reaching 100%. I don't always clear hash between analysis of different positions, but I do find by closing and restarting the program this gives a boost to the nps. Not quite sure why this happens but is prevalent with both Komodo and to a lesseer extent Houdini.
Given that (from what I understand) most people who purchase Komodo do so for analysis....seems like this should be top on Team K to look into!
kk wrote:I have noticed another quirk in longer analysis sessions in that the nodes per second does tend to reduce. This happens to both Komodo and Houdini and I wonder whether it may be tied up with hash reaching 100%. I don't always clear hash between analysis of different positions, but I do find by closing and restarting the program this gives a boost to the nps. Not quite sure why this happens but is prevalent with both Komodo and to a lesseer extent Houdini.
I'm under the impression that in endgames the nodes per second is much higher and this is due to fewer pieces on the board but also perhaps to fewer program instructions at this stage of the game for each node reached? Don't know if I am right on this but those are my thoughts to date.
The second first: yes, the fewer the pieces on the board, the quicker Komodo (and probably most programs) get. There is less to do both in move generation and in board evaluation.
I cannot think of any reason restarting Komodo might make it quicker. One idea is modern operating systems, when they allocate big batches of memory for things like hash table, often are unable to make the memory blocks contiguous. Mapping is used to make it appear that it is one block of memory, but this is slower than a real block of memory all in out place. One reason Large pages is faster. After running a while, then closing and restarting a program might help in this regard. Just a speculation. Do you see the same thing when you Clear Hash? I can look into this.
I mainly use Hiarcs Explorer for its fuss free simple user interface. It also appears to run engines a tad faster (perhaps slightly less overhead) than the chessbase interface.
I haven't been successful in finding a clear hash command in hiarcs explorer so i normally Switch engines then switch back to the engine I want to use and that often results in a 'refresh' of speed. This is only relevant after a long think time, perhaps overnight, equating to 10 to 15 hours. I may run several analyses on different positions throughout a week then suddenly find a slow down after a longer think and that is when I would try an attempt to reset things.
I have a sneaking feeling that sometimes after an overly long think time the CPU is running hotter; I always use the maximum threads i can get away with which doesn't show a speed reduction in the short term and doesn't lock the m/c up so it won't let me do anything else without an annoying delay.
I suspect temperature plays an important part in the op. system deciding at what speed the chip will run. The speed of the chip may well be controlling the possible nps, my guess here.
If temp. is a part of the problem (I only use laptops) then stopping and reloading thg program gives the system a short break and a temperature drop then on restart I would imagine a higher clock rate on the chip is restored which would show a higher nps. If so this is clearly not the fault of the program to control.
Incidentally as described in an earlier post if Komodo goes into what appears to me a long think whilst showing a high winning score for the side to move this may also occur when a score of zero is shown and a repeating position is calculated. Komodo also appears to go into non active mode when it has found a mate, so I'm not sure if it is looking for a shorter mate or has had enough and sits there.
Very occasionally I have revisited the analysis in the morning and found a messsge box telling me the program has stopped, for what reason i know not. I am running on High Sierra and the latest updates in Windows 10. Reloading the engine corrects the error.
The only other thing i would mention is that I think Komodo could do a little better at finding mates. It consistently finds mates later than Houdini 6.0.3 but does always eventually come up with a similar display even though it may be between 10 and 20 minutes later, to give a common example.
I must say move correlation between Komodo and Houdini on the same positions is high. Following a longer analysis period Komodo, Houdini and Stockfish are often showing the same candidate move and rather amazingly very close position evaluations scores too!
There are so many variables to control when doing testing between programs (hardware, threads, hash size, opening book type, time of reflection, etc) i find it very difficult to come up with a clear winner between Stockfish (latest dev version), Houdini 6.0.3 anf Komode 11.2.2!
