jackcs
New Member
never mind
Posts: 12
|
Post by jackcs on Jul 23, 2014 14:08:48 GMT -8
In sailboat racing, the best tests of a sailor's ability are the one design competitions where all the sailboats in the race are close to identical. Some even go to the extremes of exchanging boats between races. I am surprised that in the Game Playing competitions, the hardware is allowed to vary greatly. While I understand algorithms can provide exponential improvements, the games are definitely compute time limited and the compute times must be carefully managed. The bigger faster machine can beat a smarter player on an inferior machine. (So says woodyHead who runs on a 600 Mbyte free Amazon slice . LOL)
Are the advantages of faster bigger machines evident in the previous world championships?
|
|
|
Post by Sam Schreiber on Jul 23, 2014 16:41:50 GMT -8
I dunno, I won the 2013 and 2011 world championships with TurboTurtle running on my laptop, beating CadiaPlayer (which used an eight computer cluster) and Ary (which used a forty computer cluster).
|
|
|
Post by alandau on Jul 23, 2014 23:42:39 GMT -8
Given that most players write their own rules interpreters in one way or another, the difference in code performance there can more than make up for the hardware difference. Making your code run ten times as fast is even better than running it on a ten-computer cluster, because then you don't have to deal with the costs of communication and coordination.
And (as far as I can tell) even with UCT, there is a certain point where throwing more raw cycles at the algorithm gives diminishing returns, and you'll usually lose to better-tuned variants of the algorithm that converge faster.
|
|
|
Post by Andrew Rose on Aug 13, 2014 10:53:30 GMT -8
Being a sailor myself, I understand the whole one-design concept. However, GGP is still at the stage where algorithmic differences make far more difference than hardware. Sancho won IGGP14 whilst limiting itself to using half of a single 4-core (+hyperthreading) CPU.
Perhaps there'll come a day when programs will need to be submitted to run on standard hardware, but I think we're several years away from that.
|
|