I have 2 NVIDIA cards, the primary is NVIDIA GTS 450 and the secondary is Tesla C2075.
linux 3.6.9 => linux 3.7.9-2
nvidia 310 => nvidia 313.18-3
nvidia-utils 310 => nvidia-utils 313.18-1
opencl-nvidia 310 => opencl-nvidia 313.18-1
The cudaGetDeviceCount function in CUDA didn't recognize Tesla C2075,
and cudaGetDeviceProperties couldn't find it as well.
Strangely enough, Tesla C2075 still shows up in nvidia-smi and nvidia-settings,
and some programs using opencl-nvidia interface could recognize Tesla C2075 as well,
the problem is only with CUDA programs....
I tried reinstalling cuda5 and recompiling the binaries,
switching the Tesla C2075 to the primary video card,
but they didn't make any difference.
Any idea on how this could happen?
Now I just down-graded nvidia driver and Tesla C2075 shows up in CUDA programs again....
Last edited by liuexp (2013-02-27 13:06:34)
Strange enough, the issue just disappeared with the latest update,
i.e. linux-3.8.4-1 & nvidia-313.26-1