Performance Modeling Lead

Key Responsibilities

• Build and own a performance modeling framework/toolchain to evaluate AI systems across multiple levels of abstraction.
• Analyze and quantify architectural tradeoffs across compute, memory, networking, storage, and system topology.
• Develop performance models to guide decisions on: scale-up vs. scale-out architectures, interconnect and network design, memory hierarchy and system balance.
• Translate modeling outputs into clear recommendations for internal teams and external hardware vendors.
• Influence reference designs and vendor roadmaps through data-driven insights.
• Partner closely with machine learning, systems, and hardware teams to understand workload characteristics and requirements.
• Lead and grow a small team (2–3 engineers), setting technical direction and maintaining high standards for modeling rigor.
• Continuously improve modeling fidelity by validating against real system behavior and measurements.

Qualifications

• Experience owning or building performance modeling frameworks used to drive real system design decisions.
• Deep knowledge of AI/ML workloads, including training and/or inference at scale.
• Understand system-level tradeoffs across compute, memory, and networking in large-scale distributed systems.
• Comfortable working across abstraction layers—from workload behavior to hardware implementation.
• Experience using modeling (analytical or simulation) to inform architectural decisions.
• Can operate in ambiguous problem spaces and turn open-ended questions into structured analysis.
• Communicate clearly and influence both internal teams and external partners.

Preferred Skills

• Experience working with hardware vendors (ODM/JDM, silicon, networking).
• Background in data center infrastructure or hyperscale systems.
• Familiarity with accelerators (GPUs/ASICs) and interconnects (e.g., NVLink, InfiniBand, Ethernet).
• Experience influencing hardware roadmaps or reference architectures.
• Prior experience leading or mentoring engineers.