Marketing idea: CPU vendors should start using GPU vendors' practice of calling each SIMD lane as a "core"... Coming soon: Intel 16 unit, 256 core CPU.
Fun fact: Intel has a compiler project "ispc" that allows you to write CPU code in GPU SPMD sytle. The compiler automatically maps SPMD code to SIMD instructions. This programming model means auto-vectorization is not a hit or miss, it always works. https://github.com/ispc/ispc