when I started using the compile that you suggested, I started getting losses on time, so I reverted back to the tried and trusted exe that is slower. Sorry, but I did try.
How strange! It is suposed to be exactly the same version. Perhaps uMax is more sensitive to time forfeits at certain TCs. (And speeding it up at the same C is equivalent to giving the old version more time.)
Anyway, using the slow version explains why it might perform worse than Swami expected. In Chess War the 1.75x faster P2 compile was playing, and at the TC used there, had no problems.
when I started using the compile that you suggested, I started getting losses on time, so I reverted back to the tried and trusted exe that is slower. Sorry, but I did try.
How strange! It is suposed to be exactly the same version. Perhaps uMax is more sensitive to time forfeits at certain TCs. (And speeding it up at the same C is equivalent to giving the old version more time.)
Anyway, using the slow version explains why it might perform worse than Swami expected. In Chess War the 1.75x faster P2 compile was playing, and at the TC used there, had no problems.
Well the good news is that I've never had any problems at all with the slower version.
Yes, but this must be accidental. If you would run the slower version at a 1.75 times lower time control, it should behave exactly as the faster version, i.e. also produce time forfeits...
Btw, I get the occasional time forfeit in any version of uMax 4.8 or Fairy-Max. Sometimes it starts thinking about a move ~100 times longer than average. In particular at the moment where it switches off null move, because nearly all non-pawn material gets traded away. But it is almost always the losing side that forfeits in this case, so I never saw much point in doing anything about it. It should just be considered uMax' was of resigning.
Yes, I recall getting the occasional time loss with the old version, so Graham must just have been lucky. I get the same with the new version. I had wondered if it was anything to do with running through the wb2uci adaptor, but it isn't - it still happens running as native winboard under Arena. Which is the better way to run it anyway actually
How many ELO gain would you estimate with the 1.7 - 1.8 times speedup ?
AMD64x2 Dual Core
ChessGUI
128mb hash each
3-4-5 piece tablebases
Ponder off
HS-Large Remis book (limited to 12 moves)
40 moves in 38 minutes repeating (adapted for the CCRL)
2 cycles 30 rounds
