Profiler runs fine, user interface shows

GTX 1070, Ubuntu 18.04.1 LTS, CUDA 10.0, driver 410.73

I’m trying to profile an application, but selecting my main kernel and running individual analyses yield the following errors:

  • "Kernel Performance Limiter" analysis yields "Insufficient Kernel Bounds Data, The data needed to calculate compute, memory, and latency bounds for the kernel could not be collected."
  • "Kernel Latency" analysis yields "Insufficient Kernel Latency Data, The data needed to perform latency analysis for the kernel could not be collected."
  • "Kernel Compute" analysis yields "Insufficient Kernel SM Data, The data needed to perform all SM analyses for the kernel could not be collected.". Instruction Execution Counts and Floating-Point operation counts work fine.
  • "Kernel Memory" analysis yields "Insufficient Kernel Memory Data, The data needed to perform memory bandwidth analysis for the kernel could not be collected."
  • "Memory Statistics" analysis yields n/a values for bandwidth all over the place. No actual data is listed.

I’ve tried purging and reinstalling the driver with no luck. Since a new driver was released not too long ago my theory is that the update caused some kind of incompatability. However, the profiler does run without errors. I’m completely at a loss here what could be the problem, and the error messages are utterly useless at debugging the issue.

What is going on here? Am I doing anything wrong?

Small update: I have downgraded to CUDA 9.2. The profiler now shows all analysis results without issues.

Hi bartvb,

Thanks for reporting the issue. Can you please provide more details to investigate it further?

  1. Do you see any warning/error message dialog after running kernel analysis?
  2. Can you send output from Visual Profiler console?
  3. Can you send details tab data? Click on “Details” tab. Change the summary mode to normal mode. Export details data to a CSV. Please see attached screenshot for how to do this.