The SPU's are a fundamentally different architecture to an x86 been something akin to a graphics card, heavily optimised for a very particular kind of floating point calculations (somewhat similar to SSE but much more task focused).
A more likely approach is to use something like CUDA/OpenCL but again while closer in architecture the SPU's are incredibly optimised for one thing.
I think the best chance for emulating the SPUs would be with an Intel processor equipped with Iris Pro and the shared L4 cache. The latency and memory sharing issues involved with using a dedicated GPU would impose a huge bottleneck and the AMD APUs are just so weak in IPC terms.
The SPU's are a fundamentally different architecture to an x86 been something akin to a graphics card, heavily optimised for a very particular kind of floating point calculations (somewhat similar to SSE but much more task focused).
A more likely approach is to use something like CUDA/OpenCL but again while closer in architecture the SPU's are incredibly optimised for one thing.