I haven't done exact calculations yet, but I should be able to get 250-300 sol/s with RX 480 and a GCN assembler.
That's a reasonable target, however you'll find it is very difficult to write a kernel that works for gcn1.0-gcn1.2. You'll probably have to write at least 3 different kernels, and possibly more to support the different ABIs. For example the OpenCL2.0 ABI is not supported for GCN 1.0 devices under Linux (just GCN 1.1 & 1.2 devices). Neither SPIR nor HSAIL seem to expose the GDS, so the only way of writing a kernel that is portable across different AMD GPUs would be AMDIL.
http://amd-dev.wpengine.netdna-cdn.com/wordpress/media/2012/10/AMD_Intermediate_Language_(IL)_Specification_v2.pdf