SDK example deviceQuery and deviceQueryDrv failure

Accelerated Computing CUDA CUDA Programming and Performance
1

I tried to run SDK examples

deviceQuery

deviceQueryDrv

however both failed and here is the screen dump

239 fermi1: ./deviceQuery

[deviceQuery] starting…

./deviceQuery Starting…

CUDA Device Query (Runtime API) version (CUDART static linking)

cudaGetDeviceCount returned 2

→ out of memory

[deviceQuery] test results…

FAILED

Press ENTER to exit…

246 fermi1: ./deviceQueryDrv

[deviceQueryDrv] starting…

CUDA Device Query (Driver API) statically linked version

cuInit(0) returned 1

→ CUDA_ERROR_INVALID_VALUE

[deviceQueryDrv] test results…

FAILED

Press ENTER to exit…

my setup:

dual M2090 cards with redhat 5.4 system

cuda driver 290.10

cudatoolkit & SDK 4.0

my nvidia-smi result looks like

247 fermi1: nvidia-smi

Fri Dec 30 09:34:46 2011

±-----------------------------------------------------+

| NVIDIA-SMI 2.290.10 Driver Version: 290.10 |

|-------------------------------±---------------------±---------------------+

| Nb. Name | Bus Id Disp. | Volatile ECC SB / DB |

| Fan Temp Power Usage /Cap | Memory Usage | GPU Util. Compute M. |

|===============================+======================+======================|

| 0. Tesla M2090 | 0000:0D:00.0 Off | 0 0 |

| N/A N/A P12 34W / 225W | 0% 9MB / 5375MB | 0% Default |

|-------------------------------±---------------------±---------------------|

| 1. Tesla M2090 | 0000:0E:00.0 Off | 0 0 |

| N/A N/A P12 34W / 225W | 0% 9MB / 5375MB | 0% Default |

|-------------------------------±---------------------±---------------------|

| Compute processes: GPU Memory |

| GPU PID Process name Usage |

|=============================================================================|

| No running compute processes found |

±----------------------------------------------------------------------------+

anyone had similar problem ?

2

I too have this issue (getDeviceCount and cuInit both return ‘2’). I’m using CUDA 9.1 with a Tesla P100, running on Fedora 25. I’ve disabled SELinux. Rebooted and power-cycled, and nothing seems to change. I’ve contacted support, but would welcome any suggestions while I wait.

3

My issue disappeared when I replaced the GT 630 in the system with a 1050 Ti. I could then query the devices normally, both from my own code and the CUDA samples. We will try to replace the 1050 Ti with a Quadro P400, per Nviidia Chat support’s suggestion.

4

You didn’t mention anything about having a GT 630 in the system. That is a cc2.x fermi GPU, not supported by CUDA 9.1