Principal Datacenter GPU Lead Power Constrained Performance Architect

AMD AMD · Semiconductors · Austin, TX · Engineering

This role focuses on the power-constrained performance of AMD's AI/ML/HPC GPUs, specifically their Instinct accelerators. The Principal Datacenter GPU Lead Power Constrained Performance Architect will own the Perf@Power attainment from pre-silicon modeling to post-silicon correlation, involving cross-functional leadership, workload analysis, and setting PPA targets. The role requires expertise in computer architecture, power/performance modeling, and familiarity with physical design and power management techniques. Knowledge of DL/ML/LLM/MoE workloads is a bonus.

What you'd actually do

  1. Own power-constrained performance modeling, projections, implementation target setting and attainment tracking from Concept-exit (pre-si) to Product launch (post-si) for AMD Instinct AI/ML/HPC accelerators
  2. Work with business unit and product architecture to define product configurations, capabilities and P&P targets
  3. Cross functionally lead P&P attainment by working with various engineering functions including SOC architecture, System Design, IP Design, Circuit Design, Foundry team, CAD, Physical Design, Software, Power Management, Post-si validation
  4. Roll-up Power/Performance readouts, present in executive forums, motivate cross-functional teams on power efficiency improvements
  5. Mentor junior engineers and develop skillsets and methodologies in the team for important PPA domains like Power Modeling at various stages of SoC development, Cdyn/Leakage/STA/Area trade-offs and attainment, Silicon Power/Performance debug, Post Si calibration of models, Work-load power analysis

Skills

Required

  • Background in Computer Architecture, Processor/Accelerator development
  • Experience in Power/Performance projections, modeling and optimization at SoC or System level
  • 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..)
  • Familiarity with post-silicon power/performance debug, model correlation
  • Fluency in communicating with various engineering teams, business leads and executive management

Nice to have

  • Knowledge of DL/ML/LLM/MoE workloads

What the JD emphasized

  • Power Constrained Performance
  • Perf@Power attainment
  • PPA targets
  • Power/Performance