your estimate is 1 order of magnitude lower. alphago uses 10x TPU in the game with kejie, TPU has a theoretical throughput of 92 TFLOPS. that is 920 TFLOPS total. according to wikipedia, Deepblue has a throughput of 11.4 GFLOPS. that makes alphago 80843 times faster (in theory).
list seperately, I am preparing ask: how do you find published abstract? I can't find it.
F*Q
6 楼
in the game against Lee Sedol, alphago uses 10 times more computation, but is less efficient than the current version
【在 F***Q 的大作中提到】 : : your estimate is 1 order of magnitude lower. : alphago uses 10x TPU in the game with kejie, TPU has a theoretical : throughput of 92 TFLOPS. that is 920 TFLOPS total. according to wikipedia, : Deepblue has a throughput of 11.4 GFLOPS. : that makes alphago 80843 times faster (in theory).
p*1
7 楼
i found my published abstracts on google and then got the detailed information on journal website...
【在 Y*Z 的大作中提到】 : list seperately, I am preparing : ask: how do you find published abstract? : I can't find it.
a*o
8 楼
现在软件也进步了,估计任何一款中端手机都能把国际象棋大师打得满地找牙。
: your estimate is 1 order of magnitude lower.
: alphago uses 10x TPU in the game with kejie, TPU has a theoretical
: throughput of 92 TFLOPS. that is 920 TFLOPS total. according to wikipedia,
: Deepblue has a throughput of 11.4 GFLOPS.
: that makes alphago 80843 times faster (in theory).
【在 F***Q 的大作中提到】 : : in the game against Lee Sedol, alphago uses 10 times more computation, but : is less efficient than the current version