Intel's 14nm uses 1D routing, GF's 14 and TSMC's 16 use 2D. TSMC are moving to 1D at 10nm IIRC. 1D routing is lower density, but easier to fab.
Intel put a lot of effort into SRAM cell size as a process metric, which makes them look good due to the higher density BEOL metal layers (SRAM works nicely on a 1D configuration).
I do think that at some point you have to consider that the difference in transistor count per area (and performance achieved should indicate that the transistor count isn't massively off) is relevant despite the processes being different - Nvidia's 16B/600 indicates strongly a denser logic process than Intel's 9B/700 regardless of what the marketing blurb says.