Performance Modeling Engineer
EtchedJob Title
Performance Modeling Engineer
Role Summary
Work on performance modeling and hardware/software co-optimization for inference-focused accelerator systems. You'll analyze real workloads on prototype hardware, validate models against system measurements, and inform architecture and system trade-offs.
This role sits on the engineering team in San Jose and focuses on quantifying throughput, latency, and bottlenecks to guide design and optimization decisions.
Experience Level
Mid-level. The posting does not specify years of experience; candidates are evaluated on relevant modeling, architecture, and profiling experience.
Responsibilities
Key responsibilities include modeling, measurement, and collaboration with hardware and software teams.
- Develop analytical or simulation-based performance models and projections across workloads and configurations.
- Profile and analyze deep learning workloads on hardware to identify micro-architectural bottlenecks.
- Drive hardware/software co-optimization by identifying architectural features that improve performance.
- Run regressions and validate performance models against real systems and silicon.
- Inform next-generation architectural decisions by evaluating system and silicon options during design and proof-of-concept phases.
Requirements
Core qualifications and desirable additions.
- Must-have (at least one): Strong performance modeling and analysis skills; experience building analytical or simulation-based performance models.
- Must-have (at least one): Solid understanding of computer architecture and micro-architecture, especially for accelerators.
- Must-have (at least one): Experience profiling and analyzing deep learning workloads on accelerators (GPUs, TPUs, ASICs, FPGAs, or similar).
- Must-have: Solid software engineering fundamentals with attention to auditability and maintainability.
- Nice-to-have: Deep knowledge of GPU architectures and programming models such as CUDA.
- Nice-to-have: Experience mapping models to multi-chip inference systems and inference-serving optimizations for transformer architectures.
- Nice-to-have: Experience with architecture simulators and performance modeling tools (gem5, trace-driven simulators, or custom models).
- Nice-to-have: Exposure to ASIC, FPGA, or CGRA accelerator development and hardware/software co-design principles.
- Nice-to-have: Published research in computer architecture, ML systems, or hardware acceleration.
Education Requirements
Not specified.
About the Company
Company: Etched
Headquarters: San Jose, CA, United States
Etched develops purpose-built AI inference ASICs and systems optimized for transformer models, aiming to deliver significantly higher performance, lower cost, and lower latency than GPUs. The company focuses on enabling applications like real-time video generation and advanced reasoning agents, and is backed by leading investors and engineers.
