I have been wondering for all these years if both Nvidia and AMD follow some kind of sales strategy by lowering the Core clocks and shader clocks in new GPUS.
I would expect as the technology grows, things should only get better.
For example once we achieved core clock 1000MHZ and mem bandwidth of 150GB/s then for the same price or more it should only be better meaning at least 1000MHZ and 150gb/s with additional features and nothing less.
one issue was the gpu vendors were hitting a power wall with there gpu. like amd and intel did with there older cpu. intel found out you can only push silicon chip so far then it just becomes a heat issue. look at the old 200-400 ser nvidia chips and how hot they ran. the gpu chip vendor doing what intel and amd do now pack as many cores into one chip and divide the workload over the whole chip.
The work that GPU's do lends itself well to extreme parallelism. You get more performance from adding more processing units then you would from cranking up the clock rate. You can further enhance this performance with improved memory technology, either through faster memory, a wider interface, less latency or more accurate prefetch / caching. What you see the GPU manufacturers doing is constantly improving their architectures. Clock rate is determined by whatever thermal envelope they want to keep within.