Teraflops were always a horrible power metric. They're used for PR purposes because they're the biggest number available since they're just theoretical maximums. That doesn't tell you anything about what actual utilization looks like, and just as importantly, it doesn't tell you anything about the efficiency of the computation being done. Doing a bunch of math and later discarding the results shouldn't be cause for envy by anyone, but it all folds in under that nice, simple, quotable top line number. Sure, they can give you some rough idea of how two chips with an otherwise identical architecture might perform. Otherwise, they're marketing nonsense designed to grab headlines rather than inform.
Trying to convert TF numbers between architectures, even between iterations within an architectural family, is utter folly. Any engineer who got information from a console vendor quoting theoretical maximums after multiplied by some arbitrary conversion factor to make them look better would just shake her head in wonder at the stupidity of it all and look for something more meaningful. The theoretical maximum is still the theoretical maximum. It doesn't change. 12TF GCN is actually equivalent to 12TF RDNA for the purpose it was intended to measure in the first place.
What changes is how efficiently a given piece of code is going to run, and there's no fixed multiplier available that will hold true for all code. Each stage in each rendering pipeline will need to be re-evaluated to see whether decisions made because they were the fastest approach still hold true, and once you've started changing on GCN hardware compared to RDNA hardware, it becomes even less clear how to describe the relative difference in results. Yet that's all that really matters: results. What you can get on the screen in the 16 or 33ms you have to render a frame. Nobody can boil that down to "2x the power!" and have that statement hold water across a wide range of experiences.
... and yet that's what the market wants to hear. So you get cherry-picked numbers that show the best case, or the maximum theoretical, or some other nonsense. Don't worry about it. Most of us aren't going to have the time and background required to become experts and come to meaningful conclusions from marketing specs alone. Focus on the gaming experiences they enable, because then you can judge for yourself what's worth your hard earned dollars.
Yes!
