"General purpose" in this case means that the computer can run general purpose code & not highly optimized hardware specific code. The Opteron is an x86 cpu, so you can run x86 immediately. The Teslas from nVidia will probably operate using nVidias Cg or OpenCL programming languages.

If OpenCL takes off, the idea is that ones entire application could be programmed in OpenCL & then run on either x86 or on a GPU. I am guessing, for now, the Opterons will be in charge of running non-GPU optimal code & feeding data to the Teslas massively parralel stream processors.

The "general purpose" tag I think is more marketing because if you want to eek out the last drop of performance from this computer you'll probably need to write code that takes advantage of knowing the physical underlying architecture, making your code less portable.