@vgr
I’m interested in a general compute version of this proposition abstracted away from details of blockchains. Perhaps:
Logic > long-term memory > …DRAM > “HBM” > register memory
HBM in scare quotes because I don’t mean the specific hardware standard but anything at that level in the memory hierarchy.
Feels like the scalability basically tracks memory hierarchy in proportion to latency, with memoryless logic as the cheapest thing?