Besed on my experience, on dual-socket motherboards you still have one CPU device. The OS usually hides from user whether cores are located on same physical CPU or on different ones, even on NUMA machines (of course you can get detailed info, but it's not all that straightforward). And I think this behaviour is quite logical, at least on SMP machines, since there is not much difference how cores are located physically (in case of GPU, accessing other device memory is quite troublesome; on CPUs pinning all your threads to same physical CPU rarely worth worrying), so there are barely any disadvantages in melting two physical CPUs into one OpenCL device, and it has advantage of simplifying its usage.
However I've not seen this mentioned in OpenCL specification, so it's all implementation-specific, and is not guaranteed to always hold true.