|
发表于 2009-7-31 23:10:14
|
显示全部楼层
回复 #12 Youth 的帖子
官方论坛这方面的讨论也比较多,比如:
http://aqua.dwavesys.com/forum_thread.php?id=217
回帖很多,没仔细看,一个基本的积分方案是:
8-qubit: 5
16-qubit: 14
32-qubit: 40
48-qubit: 112
72-qubit: 313
96-qubit: 876
128-qubit: 2452
160-qubit: 6865
200-qubit: 19222
240-qubit: 53821
最后面又搞了个公式:
We changed the exponentiallly growing scheme (multiplying by 2.8) to a polynomial scheme which has less chance of resulting in a credit explosion.
This is the new formula for both the GPU and the CPU:
In a work unit called date-N-SM-L, where N = number of qubits, S = number of sweeps (in millions), L = Number of Trotter slices
credits = N^2 / 6 * S * (L/64)
We also take into account the density of the connections among the qubits, which will change between problem types, so two similarly named work units may get different amounts of credit.
Note that the running time increases at a rate slower than N^2, so we are giving out more credit for bigger problems.
-Kamran
但基本上都是倾向于给大的计算任务相对更多的分,项目方也对这个做了一些解释:
http://aqua.dwavesys.com/forum_thread.php?id=215 |
|