How to resolve 'Unable to get device count' GPU error

[Jan Paul]

Hi all,

just in case your run into 'Unable to get device count' GPU errors, here's what might help: run gadgetron_info (or any other GPU-using program?) once with root privileges. Thanks to Michael Hansen to point out the solution!


Details here:

Problem occurs:
user@gadgetron:~$ gadgetron_info
Gadgetron Version Info
  -- Version            : 3.9.1
  -- Git SHA1           : cb52dd92c990b9dc0fb2b81a01709f7fa785f7c7
  -- System Memory size : 48286 MB
  -- Python Support     : YES
  -- Matlab Support     : YES
  -- CUDA Support       : YES (-gencode arch=compute_20,code=sm_20;-gencode arch=compute_30,code=sm_30;-gencode arch=compute_35,code=sm_35)
    * Unable to get device count

Run as root once:
user@gadgetron:~$ sudo gadgetron_info
[sudo] password for user:
Gadgetron Version Info
  -- Version            : 3.9.1
  -- Git SHA1           : cb52dd92c990b9dc0fb2b81a01709f7fa785f7c7
  -- System Memory size : 48286 MB
  -- Python Support     : YES
  -- Matlab Support     : YES
  -- CUDA Support       : YES (-gencode arch=compute_20,code=sm_20;-gencode arch=compute_30,code=sm_30;-gencode arch=compute_35,code=sm_35)
    * Number of CUDA capable devices: 2
      - Device 0: Tesla K20c
         + CUDA Driver Version / Runtime Version: 6.5/5.5
         + CUDA Capability Major/Minor version number: 3.5
         + Total amount of global GPU memory: 4799.56 MB
      - Device 1: Quadro NVS 295
         + CUDA Driver Version / Runtime Version: 6.5/5.5
         + CUDA Capability Major/Minor version number: 1.1
         + Total amount of global GPU memory: 255.312 MB


Now it works as normal user:
user@gadgetron:~$ gadgetron_info
Gadgetron Version Info
  -- Version            : 3.9.1
  -- Git SHA1           : cb52dd92c990b9dc0fb2b81a01709f7fa785f7c7
  -- System Memory size : 48286 MB
  -- Python Support     : YES
  -- Matlab Support     : YES
  -- CUDA Support       : YES (-gencode arch=compute_20,code=sm_20;-gencode arch=compute_30,code=sm_30;-gencode arch=compute_35,code=sm_35)
    * Number of CUDA capable devices: 2
      - Device 0: Tesla K20c
         + CUDA Driver Version / Runtime Version: 6.5/5.5
         + CUDA Capability Major/Minor version number: 3.5
         + Total amount of global GPU memory: 4799.56 MB
      - Device 1: Quadro NVS 295
         + CUDA Driver Version / Runtime Version: 6.5/5.5
         + CUDA Capability Major/Minor version number: 1.1
         + Total amount of global GPU memory: 255.312 MB

(on Ubuntu 14.04)

Regards,
Jan

[Michael Hansen]

Yes, I have seen this before. This is not unique to Gadgetron. 

See for example:

https://devtalk.nvidia.com/default/topic/749939/cuda-is-not-active-unless-i-run-it-with-sudo-privillages-/