Datacenter Power and Performance Modeling Engineer

AMD AMD · Semiconductors · Santa Clara, CA · Engineering

This role focuses on optimizing the power and performance of AMD's AI/ML/HPC GPUs, involving pre-silicon power modeling, post-silicon correlation, workload analysis, and setting PPA targets. It requires collaboration with various engineering functions and scripting for automation.

What you'd actually do

  1. Pre-silicon power modeling (HW-IP, SoC, Multi-GPU nodes/clusters), Power-constrained Performance projections and Reporting
  2. Support Pre-silicon PPA target setting (for HW IPs, FW and SW)
  3. Support PPA attainment by working with various engineering functions including SOC architecture, System Design, IP Design, Foundry team, Physical Design, Software, Power Management, Post-si validation
  4. Post Si calibration of models and Power/Performance debug
  5. Analyzing key workloads for power behaviors and optimization opportunities
  6. Scripting and coding to automate routine tasks, building new tools for wider usage

Skills

Required

  • Background in Computer Architecture, Power and Performance
  • Experience in Soc Level Power/Performance projections, modeling and optimization
  • Experience in power modeling for digital and analog IPs
  • Familiarity with digital logic physical design and power management techniques (clock gating, power gating, V-F curves, p-states, on-die voltage regulation, clock integrity, etc..)
  • Experience in post-silicon power/performance debug, model correlation
  • Fluency in Excel based data analysis & charting
  • scripting experience in Ruby, Python
  • Bachelors or Masters degree in Computer Engineering/Electrical Engineering

Nice to have

  • Verilog RTL coding experience/understanding is desirable, especially low power design techniques
  • Knowledge of DL/ML/LLM workloads will be a bonus

What the JD emphasized

  • Power Constrained Performance
  • PPA targets
  • power/perf roll-ups
  • methodology innovations
  • performance and power efficiency
  • power modeling
  • Power/Performance projections
  • power management techniques
  • post-silicon power/performance debug