Error: "Unable to collect metric and event values" when profiling "Kernel Memory" on Titan V

[cckobe]

I create an default CudaRuntime project from visual studio , and profiling with visual profiler , when I click “Kernel Memory” or "Mermory Statistics’ to profiling , it ruturn error: “Unable to collect metric and event values. The application being profiled returned a non-zero return code”.I can run other analysis correct include “Kernel Latency” “Kernel Compute” ect. After profiling return error , I found that there was a DMP file generated, which indicating that my exe program crashed during the process of profiling, I open the DMP file , pausing in “cudaLaunchKernel” function , show error message “0x00007FF820F4BA7F (nvcuda.dll) (cudacode.exe.7512.dmp) has an unhandled exception at: 0xC0000005: an access violation occurs while writing to the location 0x00000000000000000000.”