Principal Architect Performance Analysis and Modeling

About the role

This role analyzes AI inference workloads to guide the development of datacenter accelerator hardware and software features. You will examine properties of multi-modal LLMs, chain-of-thought reasoning models, and video or audio generation workloads to identify functional and performance implications. Through system stack co-design, you will shape tensor core, storage, and data movement capabilities in partnership with hardware and software teams. Your work will align dataflow and collective communication approaches by collaborating closely with product, hardware design, compiler, inference server, and kernels partners. The analysis you produce will directly inform architecture decisions and implementation priorities across the full stack. You will build and maintain analytical performance models and architecture simulators to evaluate design options under realistic constraints. Your findings will support strategic tradeoff decisions for memory, compute, and interconnect in emerging datacenter inference environments. A research background with publications in top-tier venues such as ISCA, MICRO, ASPLOS, HPCA, DAC, or MLSys is a strong asset for applied analysis in this role.

Key facts

What you'll do

Investigate properties of multi-modal LLMs, chain-of-thought reasoning models, and video or audio generation workloads to surface performance bottlenecks and functional requirements.

Conduct system stack co-design that spans hardware and software to define tensor core, storage, and data movement capabilities aligned with target inference workloads.

Align dataflow and collective communication strategies through cross-functional partner engagement with product, hardware design, compiler, inference server, and kernels teams.

Develop analytical performance models and architecture simulators to explore design alternatives under realistic workload and technology constraints.

Evaluate memory, compute, and interconnect tradeoffs for emerging datacenter inference architectures using measurement, simulation, and analytical modeling.

Partner with compiler, kernel, and system teams to ensure that models and simulations reflect implementation realities and scheduling constraints.

Support design decision making by quantifying the performance impact of architectural features and workload characteristics.

Translate complex performance data into clear narratives that guide product roadmaps and investment priorities for hardware and software teams.

Continuously update your analysis approaches to reflect advances in ML model architectures, datacenter infrastructure, and simulation methodologies.

Represent performance analysis best practices in discussions with partners, documenting assumptions, methods, and results for reuse across programs.

Contribute to internal tools and frameworks that standardize how performance data is collected, modeled, and interpreted across the organization.

Mentor peers and junior analysts in rigorous modeling practices, clear communication, and effective cross-team collaboration.

Maintain a strong portfolio of analyses and models that demonstrate your ability to connect hardware capabilities to real workload behavior.

Act as a technical leader in defining performance targets, success criteria, and measurement frameworks for new datacenter inference platforms.

Requirements

The posting states a bachelor's degree requirement. BSEE with 10+ years of industry experience or MSEE with 8+ years of industry experience is required.

Academic or industry experience must demonstrate a solid grasp of computer architecture, hardware software codesign, performance modeling, and ML fundamentals such as DNNs.

Fluency in C, C++, or Python is required for implementation and experimentation.

Experience developing analytical performance models and architecture simulators for performance analysis is required to evaluate design options.

A research background with a publication record in top-tier architecture or machine learning venues such as ISCA, MICRO, ASPLOS, HPCA, DAC, or MLSys is a strong asset for applied analysis.

Initiative and self-motivation support effective partner interactions and cross-team problem solving.

Practical notes

Work is performed onsite at the Santa Clara, CA headquarters three days per week in a hybrid arrangement.

Employment is at will under equal opportunity and affirmative action policies.

Typical interview steps

Data interviews commonly include a SQL or coding exercise, a statistics question, and a case study. Candidates may be asked to design a metric, interpret an experiment, or build a small model. Some companies give a take-home analysis. Expect questions about past projects and the business impact of your work. Interviewers often evaluate how you communicate uncertainty and business impact, not only the math. Bringing a clean write-up of a past analysis to the interview is well received.

Good to know

Performance analysis for AI workloads often combines measurement, simulation, and analytical modeling to capture system behavior. Computer architects commonly collaborate across compiler, kernel, and system teams to balance hardware and software tradeoffs for inference workloads. Models and simulations support design decisions for emerging memory and compute architectures in datacenter inference environments.

Questions to ask

Good questions to ask the employer in the interview: what does success look like in the first six months, how is the team structured, what is the current biggest challenge, and how are decisions made. Asking about growth paths and the review process is also well received. Employers expect questions, and good ones show preparation.

Career growth

Data careers grow toward senior analyst, staff data scientist, or data engineering lead. Many professionals specialize in machine learning, analytics, or infrastructure. Cross-functional work with product and engineering teams becomes more important at senior levels. The field changes quickly, so continuous learning is part of the job. Professionals who can translate numbers into decisions tend to advance fastest.