lecture1.pptx

Introduction

Why have Apple’s M-series chips taken a significant lead?

• Lower TDP
• Shared memory between the CPU and GPU
• RISC vs CISC
• Near memory computing
• Hardware specialization
  • NPU, Media encode, decode
• Heterogenous computing (efficiency/performance cores)

CPU Performance Equation

$\mathrm{CPU\,Time} = \dfrac{\mathrm{Instruction\,Count} \times \mathrm{Clock\,Cycle\, per\, Instruction}}{\mathrm{Clock\,Rate}}$

Moore’s Law: Double transistor count each year

Dennard’s MOSFET Scaling: $P = \alpha \times CFV^2$

$\alpha$: transistor switching rate (relating to clock variation)

• Transistor smaller → capacitance and voltage decreases
• Increase frequency without increasing power

However, as transistors got too small, the leakage current increased, could not reduce voltage (breakdown of Moore’s Law)