missing kernel image on Jetson nano

Accelerated Computing CUDA CUDA Setup and Installation
1

This is probably a trivial question…

I have a Jetson nano up and running, and installed a finite-element package with CUDA support from LLNL known as MFEM (MFEM.org). I was able to build the package using NVCC, but when I tried to run one of the examples with the CUDA device option, I got this error:

CUDA error: (cudaGetLastError()) failed with error:
→ no kernel image is available for execution on the device

Are there additional steps that I need to take to get a working CUDA-enabled executable?

2

You have to compile the application with appropriate architecture specification for the device.

Jetson nano should be a cc5.3 device

Specifying that in your build would involve adding a switch like:

-arch=sm_53

to the nvcc compile command line.

And if the MFEM code requires features that are not present in the compute capability 5.3 device, then you will get some compile errors at that point.

3

I’ve now built everything with -arch=sm_53 and it works! Thanks1