CUDA Resource monitor how to monitor what my program is doing!

Hi all,

I’m starting to use GPU for scientific computation: I was thinking about a tool
to profile the runtime resource usage of the GPU, something like:

using n cores on a total of m cores
using n registers …
using n kb of global memory
using n kb of shared memory
using n GB/s bandwidth

A sort of the “top” command of linux :-)

Do you have any idea if such a tool exist?
Or how to build it?

thank you

All those information you can get from visual profiler (under Windows). Although it is not an online tool (you don’t see values while the program is running), but you can examine all these laster. Visual Profiler is now bundled with CUDA Toolkit. If you have the toolkit, you should have it most likely under Start -> Programs ->NVIDIA Corporation -> CUDA Toolkit -> CUDA Visual Profiler -> CUDA Visual Profiler.

Does anyone know if there is any work being done on GPU cluster monitoring tools? A GPU “top” would be a great start.

It’s something I’ve been looking into. Another set of information which would be interesting to many, I think, is to monitor and log transfer on the PCIe links. In particular in larger clusters should be helpful


  • matthias


Just started to take first steps into CUDA. So I also am looking for nice tool to monitor my GPU usage. Problem is that as I understand Visual Profiler is not available for Linux. Is there any alternatives?

Happy new year, by the way! :)

EDIT: Just found Visual Profiler in my OpenSUSE installation. Sorry for that. Still - the question stands, how is the situation in GPU resource monitoring?