Question: CPU vs GPU innovations

So, I was just wondering. Why do CPUs seem to be topping out?

Every (Intel) generation is only about 10% better on IPC than the last with the same/similar speed.

However, GPU generations make much greater leaps, like this



Why is this? Or is GPGPU just getting more research dollars than CPU development, and that trickles down to gaming cards?

CPUs are complex beasts that do a bit of everything. IPC increases come about from optimisations and/or new features. From a FPU perspective, it has far more than 10% a generation. Sandy Bridge doubled its performance compared to previous (AVX). Haswell added another 50% to that (AVX2), and Skylake about 14% again (optimisation), and we're waiting to see what AVX512 might bring when it reaches consumer CPUs. We see new instructions added to do new things faster, but the older a function is, the less likely it is to be improved upon.

GPUs on the other hand are more or less a scaling exercise. They do relatively simple tasks, but a whole lot of them at the same time. Want more power? Keep stuffing more cores in.

To write a simpler version of what mackerel said; CPUs perform serial calculations and GPUs perform parallel calculations.

GPUs are highly parallel in function. So when they add more CUDA Cores or Texture Units and ROPs with a higher core clock speed they scale much higher than a CPU.
Also GPUs just went with a massive node shrink from 28nm to 16nm FinFET, which allows higher seed Core clock because the switching distance in the transistors is shorter.

X86 CPU's work with Serial and parallel operations using FPU (flouting point unit) also integer. With a small number of cores for parallel operations CPUs don't scale well, also clock speed has not increased in a long time for serial operations.