0

I am trying to fix some error related to a SEGMENTATION FAULT. So when I try to fix the error using by step by step debugging of the code, I got couple of errors:

ERROR: cuda_trace_obj::initialize_cuda_library: Cuda initialize() returned CUDBG_ERROR_INITIALIZATION_FAILURE(20)!

ERROR: cuda_system_status_t::initialize: Error CUDBG_ERROR_UNINITIALIZED(5) getting device count

Any help or pointers regarding the above mentioned errors is appreciated.

Community
  • 1
  • 1
user1801733
  • 147
  • 1
  • 3
  • 11
  • 1
    Can your run other CUDA codes normally (e.g. an SDK sample, like deviceQuery)? Can you localize to the line of code where the sementation fault is occurring, and then paste the code leading up to that point into your question (i.e. edit your question)? I assume that getting device count is happening pretty early in your application. – Robert Crovella Nov 25 '12 at 02:38

1 Answers1

1

This error often occurs when you debug a CUDA application on a computer with a single GPU and an X11 server running.

In a single GPU system, CUDA applications can be used debugged only if no X11 server (on Linux) or no Aqua desktop manager (on Mac OS X) is running on that system.

As far as I know, only the command line debugger CUDA-GDB is able to override this restriction setting software preemption as described in the cuda-gdb documentation, but works only for devices with SM3.5 compute capability and higher.

Luca Ferraro
  • 672
  • 4
  • 12