Title The relative volumes of wires and components in a computer Author Warren D. Smith Abstract In a computer with a fixed number of ``components'' of fixed sizes and a fixed ``wiring diagram,'' what wire thicknesses will result in maximum performance? For a model of neurons, Chklovskii and Stevens argued that the ``wires'' should optimally consume $x = 3/2$ times the volume of the ``components.'' For 2D VLSI with thin sheet conductors we argue $x \to 0$ is best (albeit difficult to achieve in practice). For 2D or 3D VLSI with cylindrical wires, $x \to \infty$ is best if RC delays in wires dominate; $x \to 0$ is best if source resistances dominate. For fiber optics and superconducting transmission lines $x \to 0$ is best. For a model of the ``ultimate'' computer achieving thermodynamic limits on inter-component information fluxes, the same mathematics as Chklovskii and Stevens's neuron model re-occurs for a different reason, so that again $x = 3/2$ seems best. \end{Abstract} Keywords Neurons, fiber optics, superconducting transmission lines, computer architecture, VLSI theory, thermodynamic limits on information flux.