Principal Modeling Engineer

Principal Performance Modeling Architect - North Carolina - AI Accelerator Startup

An innovative AI hardware startup developing next-generation AI accelerator technology are looking for a Principal Performance Modeling Architect to lead the development of a system-level performance modeling platform that influences silicon architecture decisions before tape-out.

This is a highly technical, hands-on individual contributor role where you'll build and extend analytical performance models, evaluate AI hardware architectures, and help shape the future of advanced AI compute systems.

• Own and develop a system-level performance modeling platform for AI accelerator and cluster architectures.
• Build analytical models to predict performance, power, energy efficiency, and cost across different hardware designs.
• Expand support beyond LLM inference to training, multimodal, and vision workloads.
• Work closely with architecture teams to evaluate new silicon concepts and guide design decisions using simulation-backed analysis.
• Validate models against real-world AI frameworks and continuously improve accuracy.
• Improve platform scalability, automation, and maintainability.
• Provide technical guidance to junior engineers and collaborate with customers and partners where required.

• 5+ years' experience in performance modeling, computer architecture, or systems performance engineering.
• Strong understanding of AI accelerators, GPUs, HPC systems, memory hierarchies, and interconnects.
• Experience developing analytical or first-principles performance models rather than purely benchmarking hardware.
• Knowledge of AI inference and training workloads and how they utilise compute, memory, and networking resources.
• Strong Python development skills with experience building maintainable engineering tools.
• Ability to analyse complex simulation results and influence architecture decisions through data-driven insights.
• Excellent communication skills with the ability to explain technical concepts clearly.