CUDA can't recognize GPUs after reboot Linux

I have 4 tesla k20m on my computer.
And I used cuda 5.5 with Red Hat linux server 6.5 and it worked well.
But today, from right after i reboot my computer, there are errors code=1 (cudaErrorInvalidDevice) on cuda samples.
(many samples have same problem to get device, get device property, or set device)
What’s wrong with it?
And I checked GPU status on my linux, and even os can’t recognize GPUs…
Although I re-installed graphic driver, it didn’t make it.
Is there anyone that had same problem with me?
Any comments, advices will helpful.

what are the results of:

nvidia-smi -a

run as a root user?

Also provide the results of:

lspci -v |grep -i nvidia


dmesg |grep NVRM