I find it hard to take these events seriously while they're still BYO hardware. It's a competition of physical resources, not a competition for the best software. Compare https://en.wikipedia.org/wiki/Top_Chess ... ampionship ; more discussion at https://www.reddit.com/r/chess/comments ... c_and_the/ and https://chess.stackexchange.com/questio ... each-other

That does make it difficult/impossible to judge the software.

HOWEVER ... developments in hardware are not irrelevant. these programs are using neural nets, and the neural nets are emulated by programs running on processors not originally designed for the purpose. Development of hardware more directly intended for neural netes could make an order of magnitude difference.

I'd think that a FAIR was to class machines for this sort of competition might be power consumption. We could have say a 100 watt class, a 250 watt class, an unlimited class, etc. I'm not arguing for specific wattages, just the concept. Though I do like the idea of one in the neighborhood of 100 watts (because comparable to human power)

I think Mike is likely right. While I don't know know if this is true of top Go AIs specifically, in AI in general there are some programs designed to work optimally with custom processor chips (I think this is Google's approach?) and many that use standard Nvidia GPUs. There are also differences in how AI programs take advantage of parallelization (the ability to run multiple computations simultaneously) that can reflect hardware design. You wouldn't want to dictate all competitors use the same hardware if you'd get one winner under one hardware standard and a different winner under another hardware standard.

Are there any results available?

_________________
www.boardnstones.com