Principal Modeling Architect - DC GPU
AMD · San Jose, CA · 2 wk ago
On-siteEngineeringFull-time
Key Responsibilities
- Develop and refine workload modeling frameworks to characterize and project performance, scalability, and resource utilization for AI/ML, HPC, and data analytics workloads.
- Analyze emerging model architectures (e.g., LLMs, transformer variants, graph neural networks), datatypes, and scaling methodologies to anticipate future platform requirements.
- Collaborate with architecture, silicon design, software, and performance engineering teams to translate workload insights into platform-level technical requirements.
- Lead benchmarking, profiling, and simulation efforts to validate architectural assumptions and guide design trade-offs.
- Produce detailed workload characterization reports, performance projections, and sensitivity analyses to inform platform strategy and technical decision-making.
Required Qualifications
- 12+ years of experience in workload modeling, performance engineering, system architecture, or related technical domains.
- Demonstrated expertise in modeling and analyzing AI/ML, HPC, or large-scale data analytics workloads on GPU or accelerator platforms.
- Deep understanding of performance modeling methodologies, benchmarking tools, simulation environments, and workload characterization techniques.
- Experience collaborating across hardware, software, and system engineering teams to drive workload-informed architectural decisions.
- Strong analytical, communication, and technical writing skills; ability to synthesize complex data into actionable insights.
- Advanced degree in Computer Science, Electrical Engineering, or related field preferred.
Preferred Qualifications
- Experience with ROCm, CUDA, or other GPU programming frameworks.
- Familiarity with compiler/runtime systems, kernel libraries, and developer tooling for AI/ML workloads.
- Track record of publishing workload analysis or performance modeling research in peer-reviewed venues.
- Experience engaging with hyperscalers, CSPs, or large enterprise customers on workload deployment and optimization.