Moderators: hgm, Rebel, chrisw
Laskos wrote:Houdini is cheating by launching dormant processes during its turn, which activate during opponent's turn. I checked that on various hardware, always the same. Just look at how sloppy Houdini plays without cheating. Rybka 4 is by far the strongest engine (btw IvanHoe is also cheating, Robbolito is massively cheating, soon Stockfish will cheat too).Post subject: What is strongest Rybka 4? -- and is Houdini cheating ?
Kai
The poor Rybka fanboys made tons of testing to find such settings and it all ended with massive failure....Houdini wrote:Charles,
It is very much the expected behaviour that the engine immediately activates all the threads when it gets a "go" instruction.
It is equally expected that in rapid 0+4s matches Houdini 1.5 convincingly beats Rybka 4 or Ivanhoe, 75 Elo points difference is about correct in my experience. I'm not aware of any Rybka 4 settings that would signficantly improve its match results against Houdini 1.5.
Cheers,
Robert
Why,aren't we having funzullil wrote:Seems like a good moment to lock this thread.
Thanks for your reply.. Yes, it does look like expected result -- though I am getting +120 elo at least. I will post the results when its complete just a 20 more games to go.Houdini wrote:Charles,
It is very much the expected behaviour that the engine immediately activates all the threads when it gets a "go" instruction.
It is equally expected that in rapid 0+4s matches Houdini 1.5 convincingly beats Rybka 4 or Ivanhoe, 75 Elo points difference is about correct in my experience. I'm not aware of any Rybka 4 settings that would signficantly improve its match results against Houdini 1.5.
Cheers,
Robert
The Elo difference I quoted was for the 64-bit version, with 32-bit the Elo difference is probably somewhat larger - Houdini is very well optimized for 32-bit, Rybka 4 not so much.frcha wrote:Thanks for your reply.. Yes, it does look like expected result -- though I am getting +120 elo at least. I will post the results when its complete just a 20 more games to go.
If testing was done on two separate computers and still maintained the same elo difference between Houdini and Rybka then that would suggest there is no cheating going on.bob wrote:If you use the same machine to play both engines, there is lots of room for mischief. When I was a grad student, we had an othello tournament for a class project. I did a parallel search, and used N threads on a single-core machine. If I set N = 10, where my opponent did not do a parallel search, I would get about 90% of available computing cycles, or I ran about 10x faster than he did. And this was ponder=on (for those [not many] that did pondering). So yes, you can play games.frcha wrote:I am having some strange results in my engine vs engine tournaments without houdini 1.5 - rybka 4 seems very strong -- it is about even to ivanhoe over a long series of games -- at time control of 0+4s I use this control to offset any possible bugs in arena and also to possibly make R4 stronger since supposedly you need to change their settings to make it perform better.
Are there known proven to be stronger parameters for R4 in testing eng vs eng for various types of time controls?
Now when I started playing Houdini vs R4 -- I am getting very lopsided results -- it totally demolishing R4 up to 130elos!!! Its not doing so well against the latest stockfish only up by 50 elo. I am playing a 100 game gauntlet match of H vs R4 and H vs SF 2.01
Its ponder = off and 0+4s and with the silver suite.
Is it possible for an engine to gain an advantage by simply being able to grab most of the cpu and hold on for just a tad longer? I notice from the task manager than when its houdini's turn to play - it grabs the cpu usage to 99 very quickly while Rybka 4 each process takes a little longer to reach 25 (4 cpu) .
Its probably nothing but can anyone chime in with this? Can an engine actually gain elo from these types of matches by taking hold of the cpu quickly ?
If you do ponder = off, I would be temped to do a normal search, then spawn 100 threads while waiting on my opponent, giving him 1% of the CPU.
Linux is a bit smarter, because of thread groups, but it would still let me steal a lot of time from my opponent, which would hurt its speed and bias the results.
With all the copying going on, nothing would surprise me with respect to this group of "cousins" at the top of the list. you just get what you get...
There is no cheating going on.Werewolf wrote:If testing was done on two separate computers and still maintained the same elo difference between Houdini and Rybka then that would suggest there is no cheating going on.
Hey man !Laskos wrote:
Rybka 4 is by far the strongest engine (btw IvanHoe is also cheating, Robbolito is massively cheating, soon Stockfish will cheat too).
Kai
S 