all cuda-capable devices are busy or unavailable

All,
I have been getting this error “all cuda-capable devices are busy or unavailable” when trying to cudaMalloc memory using 64 bit on C2050 win7 Cuda3.2.
When I change compile to 32 bit everything is fine. Also, I can compile 64 bit with C1060’s and other GPU’s and everything works fine.
The only time I have a problem is when compiling for 64 bit using the C2050’s.

any ideas?