"out of memory" for all cuda function all

I’ve M2090 cards with redhat 5.4 system
and installed driver 290.10
with cudatoolkit4.0

however, for even the simplest cuda code.
whatever cuda function I use

cudasetdevice, cudamalloc, cudagetdevicecount, etc

it always return error code 2
which means “out of memory”

and I tried google it, didn’t find anything similar

anyone had this problem before?

is this a system setup problem or something else?