Tracking down non-coalesced events with profiler

I’m using CUDA Visual Profiler ver 1.1.5 to profile my applications, my GPU is C1060. But in the Profiler Counter page, I cannot select gld_uncoalesced and gst_coalesced. I don’t know whether it’s a hardware limitation or limitation of the version of VisualProfiler I’m using. If it’s limited by HW, which HW can support this counter? Thanks!

Because GT200 (which is what your C1060 has) will try to partially coalesce loads/stores, there’s really no such thing as an uncoalesced load/store. Instead, use the transaction size counters. Look at Chapter 5 in the programming guide for more info on how GT200 determines transaction size, etc.