what's the benefit of capability1.3 in coalesced memory access, compared to capability1.1

I’m new in cuda and,

My graphic card is gtx285 and the capability is 1.3(I think so, because I tried devicequery and it said the major reversion is 1 and the minor reversion is 3)

So if I want a coalesced memory access, in 1.3, should the pattern that the alignment of the data address and the same index of the threadidx and data relative-position be matched?

and how about the bank?


There is no coalesced and non-coalesced anymore with CC 1.3, it is now the number of memory transactions that counts. Having said that: if your memory accesses are coalesced on 1.0/1.1 hardware, then you also have the minimum amount of memory transactions