I needed to test my CUDA application on a low-end machine so I bought a Jetway Mini-TOP “nettop” machine.
Like most of the dual-core Atom+ION2 nettops that are available, the GPU is connected via a single PCIe 1.0a lane so the speeds you see are only a couple times faster than old PCI.
But, it seems to get the job done and once onboard the GPU, the 512MB of DDR3 gets busy. :D
I threw up some pics, screen caps and cut-and-pasted output from a few of the standard CUDA pseudo-benchmarks on my blog.
Let me know if you have any questions or other results you would like to see.