There's AMD pulling ahead, but that doesn't answer why Maxwell is so much farther ahead than Kepler.
Before we get too far down this road, I want to make clear that I am only offering other possibilities. I believe very strongly that these factors are at play here, each to some degree. That I am fairly certain that the singular stance that the entire situation is due to nvidia abandoning kepler is just way to shortsighted. It is lacking meaningful thought on the matter.
You ask why Maxwell is so much further. I do believe you are seriously asking this question but I dont think it is really so hard you cant imagine scenarios yourself. The most obvious comes to light simply by looking at any new architecture almost a year later. From Fermi to Tahiti, they all improved over their predecessors, widening the gap. Every other generation though, we called it driver improvements. It is not a new concept, it used to be the absolute expectation. What is so strange about Maxwell? We should all have expected that performance would improve the longer the driver teams have to work with it. It seems now, with this architecture people just believe that there is no improvement to be had for some strange reason. That instead of Maxwell improving, people see it as nvidia sabotaging kepler.
I think the outlier here is how the performance on GCN has gone up in this same time period. And this may be the factor at play that is causing the fuzzy vision. See, we often have Nvidia and AMD launching new architectures around the same time. This sort of brings us a brand new clean slate. But this time, things are different. Even more confusing is that AMD has steadily been gaining performance due to the factors i wrote about in my previous post (their on going driver jumps and consoles having GCN graphics, etc).
AMDs improvements are undeniable, just look at the 980 launch articles to the more recent ones. The 290x has gained ground on the 980. It is clear as day, they have closed the gap. But, i am also asking you to think about another element at play here. One that makes things a little more complex. I propose that Nvidia has been improving maxwell performance, at the same time. That their drivers are becoming mature, just like every other generation.
So, if nvidia has been improving their maxwell performance and AMD has still been able to close the gap at the same time...........
It is not so strange that kepler is being left behind.
When you were little and riding in a car on the interstate, do you remember watching vehicles appear to be going backwards as the car you were in passed them up?
It is a matter of perspective.
Q) So why is kepler not keeping up?
A) Nvidia had many years to extract the most they could out of kepler. There is no choice in the matter, Maxwell had/has more to gain simply being a younger architecture.
Q) But AMD still seems to be making progress with GCN
--ah yes, but also we have to realize that the next gen consoles are GCN. We also all know that AMDs architectures were way more powerful just looking at the specs. Even in gflops, AMD built some really beefy chips. I am suggesting that Nvidia was getting more out of kepler early on. They were able to extract and utilize them to their fullest. AMD has steadily been improving their performance and over the years we have seen multiple large jumps in big driver releases, as well as smaller gains with betas.
AMD is heavily vested in GCN as it will still serve them for yrs to come and they will continue to improve and extract performance.