The von Neumann machine is the past, the AI PC the transition, the Noeton the claim. Eight dimensions, row by row — click a column header to focus one machine.
Organizing center
CPU & instruction stream
Still the CPU; NPU bolted on
The memory-resident model
Primary compute
General cores (OoO/speculative)
GPU/NPU-accelerated subroutines
Deterministic tensor dataflow fabric
Role of memory
Passive store, hauled by the word
A bigger passive store (HBM)
First-class: PIM + pooled residence
Operating system
Processes / files / syscalls
Same, plus an AI assistant layer
Model-OS: LLM kernel
Unit of interaction
Commands on data
Commands, occasionally prompts
Intent against context
Role of the network
Optional peripheral
A pipe to cloud services
Part of the machine: weight pools
Latency character
Cache-luck statistics
Batch jitter; p99 runs away
Compile-time scheduled; contractual
How it upgrades
New CPU / more DIMMs
A bigger GPU
Swap weight pools in the network, invisibly
The AI-PC column summarizes today's mainstream "general machine + NPU" form; comparison follows Chapter 7 of the paper.