After the first half of the bullet 20"+1" tournament, 8 threads, hash=4096, Ponder=ON, with Move Overhead increased from 30 to 75 for all the fish and to 1 sec for Komodo 11.2,
Gusev wrote:After the first half of the bullet 20"+1" tournament, 8 threads, hash=4096, Ponder=ON, with Move Overhead increased from 30 to 75 for all the fish and to 1 sec for Komodo 11.2,
????? Why would you use 75 ms overhead for SF, and 1000 for Komodo? You are having Komodo play with no increment that way! No wonder it is near last place. Normally the default 30 should be enough, but if you need to raise SF to 75, you should do the same for Komodo.
Gusev wrote:Hi Larry, I tried that setting before, K just keeps forfeiting on time under these conditions in Arena, no matter what.
Well, I don't use Arena so I don't know what's wrong, but on other GUIs Komodo doesn't normally need more Overhead than Stockfish. Try 100 or even 200 if necessary, but if you need to use 1000 or even 500 you might as well leave it out of the tournament. Playing 20" sudden death vs. 20" plus 1" increment is an enormous handicap; I'm surprised Komodo isn't dead last.
p.s If you are running 8 threads on a four core machine, using hyperthreading, that is probably the problem. It should be fine to use 6, maybe even 7, but using all 8 weakens all the engines and causes problems like this. I always run Komodo (or Stockfish or Houdini) on six threads on my i7 quad laptop. Kai did some studies that indicated that this was best, and I have confirmed his findings.
It is even worse than that..... he is using 8 threads with ponder on... so god only knows what is going on with the two engines fighting over priority....
jhellis3 wrote:It is even worse than that..... he is using 8 threads with ponder on... so god only knows what is going on with the two engines fighting over priority....
Exactly! This is a load that shows 100% of CPU utilization. Terrible. And this is how this extreme bullet contest ended. So, I suppose, it's God and Robert Houdart.
.... You don't even understand why what you did is wrong. I guess there is no hope for a productive conversation. Ah well, carry on with your "tests"....
There is nothing wrong with boundary testing. We take an edge case (100% CPU load), we test it. Some engines handle it more gracefully than others. There were no crashes in this tournament, mind you.
jhellis3 wrote:.... You don't even understand why what you did is wrong. I guess there is no hope for a productive conversation. Ah well, carry on with your "tests"....
Gusev wrote:There is nothing wrong with boundary testing. We take an edge case (100% CPU load), we test it. Some engines handle it more gracefully than others. There were no crashes in this tournament, mind you.
jhellis3 wrote:.... You don't even understand why what you did is wrong. I guess there is no hope for a productive conversation. Ah well, carry on with your "tests"....
I don't understand the purpose of the test. What was learnt here?
There is nothing wrong with boundary testing. We take an edge case (100% CPU load), we test it. Some engines handle it more gracefully than others. There were no crashes in this tournament, mind you.
Nope, you missed the point yet again... attempt number 3 on why the results are meaningless?