It's an oversimplification to say that Moore's law is over. Some important metrics have stopped improving. Others are still going, for now.
It's an oversimplification to say that Moore's law is over. Some important metrics have stopped improving. Others are still going, for now.
It's not over, but we're seeing trade-offs that we didn't see in the past decades, like clock rate stagnation and exotic cooling but most importantly, architectural changes.
Architectural changes present the biggest difficulty IMO because they require a change of mindset from the user's perspective. Multi-core was supposed to be an easy transition but it's still underutilized, GPUs for general purpose computing is doing well, but it's based on a popular 20 year old technology (and it's still not as accessible as CPU programming).
However the move to heterogeneous platforms will require substantial effort from a developer's perspective (however good the tooling might be to abstract things). Programming a CPU is one thing but programming a CPU with embedded GPUs,FPGAs,AI dies, etc. will certainly move a lot of people out of their comfort zone
An exponential function is many order of magnitude off of the reality.
IPC increase are today more significative than clock increase, but most of the computational increases nowadays come from scalability innovations like SPARK.
But it is true that some few specialised ASICS make big leaps.