Feels like most people now think about Moore’s Law in terms of effective compute, not just transistor density. If so, Moore’s Law seems far from dead!
Conversation
Replying to
There were always valuable non-domain specific aspects of Moore's Law. Yet even it's core tenet seems to have transformed due to the common "currency" in computing shifting from transistors.
2
Replying to
With GPUs you can kinda keep making them bigger too although yields can go down a little.
Replying to
If you design something you want to sell lots of it so it needs to be better but not so good that nobody can get it's full benefit. So Moore's law becomes a self fulfilling prophesy. With each designer for CPU, RAM, manufacturing etc planning to beat current speed by just enough
Replying to
Almost nothing to do with Moore I would say, more to do with computer and system architecture
1
5
Replying to
Absolutely! I think 's work is pointing towards even greater improvements pointed towards differentiable ML that's also super interesting.
2
Replying to
If Moore’s Law == performance then it died 15 years ago (when the Dennard scaling died) for tons of problems that cannot be parallelized and single threaded perf is all that matters to solve them faster. The progress in this field was almost non-existent in the last decade.