Originally Posted by AlphaC
Keep in mind Pascal has 128 CUDA per SMM similar to Maxwell , plus it's about 40% more boost clockspeed on Pascal (~ 2 to 2.1GHz vs 1.5GHz).
Without the clockspeed and color compression techniques I don't think a GTX 1060 6GB would be just short of a GTX 980.
It looks to be
GTX 750 ---> GT 1030 (384 CUDA) ... notice how this should have been on par with GTX 750 Ti
GTX 950 ---> GTX 1050 (640 CUDA)
GTX 960 ---> GTX 1050 Ti (768 CUDA) ... with 4GB VRAM similar to some GTX 960 variants , should be 6 SMM (it is)
GTX 970 ---> GTX 1060 3GB (1152 CUDA) ... similar to the 3.5GB fiasco , should be 10 SMM
GTX 980 ---> GTX 1060 6GB (1280 CUDA) ... a VRAM upgrade but falls short of a GTX 980 , really should be 12 SMM
GTX 980 Ti ---> in between GTX 1070 8GB (1920 CUDA) and GTX 1070 Ti 8GB (2432 CUDA) , really should be 16 SMM
Thanks for the update. Although some might not agree about the 980 TI. However since TPU's breaking news the 1070 TI will have 2432 CCs with 152 TMUs!!! That's 8 less then the 1080!
I would like to see the impact of performance when regards to the 980 TI with 96 ROPS vs 1070 TI with 64 ROPS. At the same clock speeds. Sure, there are other implications as you've suggested.
It's still interesting none the less if we've reached moore's law when it comes to ROPS.