Skymont little cores have 4x 128-bit execution. They could quadruple-pump.

But looks more like they're giving up on people writing code for wide vectors, instead settling on trying to make the existing code faster.