I finally found a die shot of an RTX GPU that wasn't supplied by NVIDIA. Which means time to go die comparing!

Initial guess at the moment is between a Pascal SM and a Turing SM, both the tensor and RT cores increased the area of the SM by about 1.88 times. However I can't figure out what part is actually the tensor and RT cores, since organization of everything is appears different between the two.

Look forward to more soon (tm)