Profiler and un-/coalesced memory access


I want to profile my CUDA app and I’m especially interested in uncoalesced memory accesses.
But the checkboxes under Profiler Counters are deactivated. I’m using version 2.3.10.
What can I do to activate them?

Thanks for your help!

For CUDA Devices of capability 1.3 these un/coalesced access patterns dont apply anymore. You wanna look at the number of the different memory fetches. As far as I know its better to have the large ones, since that means that you use the full bandwidth (512bit=64byte=8floats=4doubles at once).

best regards