7

I was about to spend some time researching this, but realized someone might have already looked into it; if not, I may just answer my own question in a few hours.

Scid is pegging one core of the CPU, possibly trying to keep up with input from Stockfish? Stockfish is using 0% CPU, possibly blocking while piping output to Scid. That is just what I suspect is happening--haven't confirmed. In any case, stockfish takes a minute or two to get to 2 depth in analysis--usually after a few seconds I'm to 15 depth at least, so it's orders of magnitude slower.

I didn't change any of my UCI settings, just simply upgraded the executable. Same thing happens on Mac and Windows, so it's probably not platform specific, but something related to the UCI interface?

I have hash size set to 1024MB, "use sleeping threads" = true, and multiple variations set quite high (100 or so), to see all variations scored and searched, which helps in analysis even though it slows down the depth search. The rest of the settings are default (I'm fairly certain).

Improve this question
4
  • Update: I don't think this is related to Scid, as I've tested in another UCI GUI and it runs even slower.
    – Eve Freeman
    Jun 4, 2012 at 19:07
  • Perhaps related, perhaps not: support.stockfishchess.com/discussions/questions/…
    – ETD
    Jun 4, 2012 at 22:09
  • Yeah, I saw that. I don't think that's quite the order of magnitude difference I'm seeing. Something is wrong. I ran the benchmark on the two, and the results seem comparable. I'm not sure why the UCI interface is so slow.
    – Eve Freeman
    Jun 4, 2012 at 22:22
  • Yeah, I figured you'd probably have come across that already.
    – ETD
    Jun 4, 2012 at 22:25

1 Answer 1

Reset to default
5

So, I hashed this out some in a support thread on the stockfish forum.

Basically, you should skip 2.2.2 and compile the latest from github (or just continue to use 2.1.1). I think I'll try this new code for a while to see how it does. The latest code from github works great in chesstempo's UCI connector. It's still a little slow out of the gates in Scid (and Scid acts like it's hanging and doesn't respond to UI commands), but it speeds up as it gets deeper in the search, so it's bearable after 30 seconds or so. I had trouble getting multipv (multiple variations) to work at high numbers, though.

I think Scid should not block the input from the UCI interface--buffering more, if need be. Maybe I'll take a look at the Scid code later to see if I can improve it.

Improve this answer
2
  • I can't believe that they actually released something worse, it takes less than 10 hours to fully test one version vs the other with an average computer
    – ajax333221
    Oct 11, 2012 at 0:26
  • It's not actually worse, in terms of performance. It's just the amount of output it pushes out causes it to act slow in some of the GUIs, especially with a high multipv setting.
    – Eve Freeman
    Oct 11, 2012 at 22:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.