CareerPlan

Principal AI Performance Architect

United States, California, Mountain ViewFull-time2026-11-17 → 2026-11-19

Core

Analyze Frontier AI training and inference workloads to identify performance bottlenecks, develop analytical models and simulators, and translate insights into architecture requirements for large-scale AI systems.

Role type

Principal AI Performance Architect

Builds

Analytical models, simulators, and data-analysis tools for AI system performance

Domain

AI Systems Performance / Computer Architecture / High-Performance Computing

Deliverable

production ML models

Required skills

AI systems performance analysis, computer architecture, performance modeling, high-performance computing, distributed systems, accelerator software, Python, C++, CUDA, Triton, ROCm/HIP, transformer-based models, large language model training and inference, attention mechanisms, KV-cache behavior, mixture-of-experts, speculative decoding, distributed execution strategies, accelerator architecture, memory hierarchies, scale-up and scale-out networks, kernel optimization, root-cause analysis, data visualization

Preferred skills

None stated

Technologies

Python, C++, CUDA, Triton, ROCm, HIP

Responsibilities

Analyze Frontier training and inference workloads to identify key performance requirements and bottlenecks; Develop and improve analytical models, simulators, and data-analysis tools for AI system performance; Evaluate trade-offs in performance, scalability, utilization, capacity, and cost across accelerator, memory, network, and software configurations; Study hardware-software interactions in large-scale AI systems and translate workload insights into architecture requirements; Work with MSI and partner teams to evaluate hardware-software co-design options for training and inference; Develop and run targeted microbenchmarks on silicon to measure compute, memory, communication, kernel, and synchronization performance; Compare model predictions with silicon and end-to-end workload measurements, identify gaps, and improve model accuracy; Communicate findings and recommendations through clear data visualizations, technical reports, and architecture decision materials.

Seniority

Principal, hands-on IC