CPU Performance Architect

About the role

You will serve as a performance architect at a semiconductor company, owning the methodology and the models that predict how fast the processor will be before it exists. Performance architects are the program's source of truth on performance: their models justify the architecture, their projections set customer expectations, and their correlation work after silicon teaches the organization what to do better next time. You will build the modeling infrastructure, run the studies that decide the architecture, and stand behind the numbers the product is sold on.

What you will achieve

• Build, calibrate, and maintain the cycle-accurate performance model for a CPU program, achieving projection accuracy within 3 percent of silicon across the benchmark suite by the time the product ships.
• Run the design-space exploration studies that select the final microarchitecture configuration, quantifying performance, power, and area for each option and recommending the configuration that best serves product goals.
• Establish the performance sign-off methodology for program milestones, defining what gets measured, what accuracy is required, and what evidence justifies proceeding to the next phase.
• Characterize customer and market workloads in depth, from industry benchmarks to proprietary datacenter traces, ensuring the architecture is optimized for real usage rather than synthetic scores.
• Lead post-silicon performance correlation, identifying exactly where the model diverged from silicon and feeding those lessons into modeling methodology improvements for the next program.
• Present performance results and tradeoff recommendations to architecture leadership and executives with clarity, honesty about uncertainty, and actionable conclusions.

What you will bring

Must-haves

• 5 to 8 years of experience in CPU performance modeling, architecture, or performance analysis with shipped products.
• Deep expertise with cycle-accurate simulators such as gem5 and the discipline of calibrating models against real hardware.
• Strong workload analysis skills spanning benchmarks, application profiling, and fleet or datacenter trace analysis.
• Rigorous quantitative habits: you quantify uncertainty, avoid overfitting models to benchmarks, and stress-test conclusions.
• Fluency in at least one major ISA and a strong understanding of how ISA features translate to performance and complexity.
• Experience presenting performance data and recommendations to senior technical leadership.
• Intellectual honesty about model limitations; you would rather report an uncomfortable truth than a flattering fiction.

Nice-to-haves

• Experience with power and thermal modeling integrated with performance analysis.
• Knowledge of ML-assisted modeling techniques or statistical performance projection methods.
• Background in performance analysis of GPUs, accelerators, or heterogeneous systems.