I was under the impression that Moore’s law was still alive and well.
Obviously game cost can’t sustain exponential growth if the market doesn’t grow exponentially (potential buyers times what they are able/willing to pay).
Nope! Quantum tunneling has made it so that the size of transistors can only shrink so far before electrons start essentially teleporting one from the other. AFAIK we might be able to get to ~3nm lithography (TSMC is shooting for it but beyond by 2020 or 2023, IBM apparently already had 5 or 7nm (forget which) lithography but hasn't published it yet), but beyond that the future is much less clear.
Shows little tendency to flatten. What I assume will happen when nVidia wants 10k stream processors on the same lithography as 5k are today, is that they will have to double the die area and struggle with yield instead? (Assuming the die can grow without communication problems)
I think it's some of each? There does appear to be a slight plateau towards the end, and the graph does seem to bear some similarity to a sigmoid function[1]. I know it's more theory, though, I saw an article in nature about it a while ago.
Obviously game cost can’t sustain exponential growth if the market doesn’t grow exponentially (potential buyers times what they are able/willing to pay).