That's not an answer to the question, that's you telling me that I don't want an answer to the question, and I'll be the decider of my own motivations, thank you. If I wasn't interested in hearing your view on it, I certainly wouldn't have wasted my time asking.
Your question seemed rhetorical. For your consideration, first I'll submit the GPU. First there was the dumb, often segmented framebuffer. Then there was some fixed 2D functionality in EGA and VGA (that few people ever explored). Then there was the fixed function 3D accelerator. Which became ever more powerful until eventually, multiple texture pipelines just wasn't enough and T&L was added. Then shaders and before you know it, the massively parallel turing-complete high-performance processing cluster that is the modern GPU was born. Virtually all of these represent different eras in graphics processing technology and were the result of numerous innovations over the previous era. And it's still going on. It's just that most people simply don't appreciate what is changing at the machine level behind their APIs. Instead, they see prettier graphics but don't realize the revolutions in hardware that have enabled it.
You can pick any component in your typical PC and see the same development. Many times, there's a complete shift in the way things work in order to overcome some fundamental limitation. In the distant past, people would laugh at the notion of using serial interfaces between system components on a mainboard (with a few esoteric transputer systems excluded). Then, as bus speeds reach a certain level, people realised that shovelling increasingly wide machine words around parallel tracks on a motherboard is both space inefficient and prone to error as the speed goes up. Simply put, the length of individual tracks becomes problematic for signal propagation. Revising everything to use multiple high speed serial links and point-to-point transfer, instead of wide transfers is an extremely clever solution that we would have scoffed at 20 years ago.
To think of all these changes as mere evolution of the same old 8086 architecture (ie, nothing new, just faster) is nothing short of disingenuous. It takes a lot to deliver all that brute force and a lot of changes to keep delivering it.
I'd say the only thing that hasn't really kept up is the software. For example, we now have many-core systems capable of dramatic throughput but lots of applications with poor concurrency support, incapable of realizing their potential. OK, sometimes that's just the way it goes as not everything can be parallelised, but we're still lagging behind in terms of how we go about designing software for such scalability in the first place.