Performance Architect, AI HW

Job Title

Performance Architect, AI HW

Role Summary

Model, analyze, and optimize real AI workloads on the Tensix compute fabric to guide hardware feature decisions and improve measurable performance, power, and area (PPA) outcomes. The role sits at the intersection of architecture, software, and RTL and works with compiler and runtime teams to validate design trade-offs across single- and multi-node systems.

Experience Level

Open to multiple experience levels; candidates will be assessed during interviews and offers will align to the appropriate level.

Responsibilities

The role focuses on workload-driven performance engineering and hardware–software co-design. Key responsibilities include:

• Benchmark and analyze complex AI workloads on single-node and multi-node hardware to identify bottlenecks and opportunities.
• Develop and maintain performance models, simulators, and micro-benchmark suites for feature evaluation and design optimization.
• Perform detailed PPA (Performance, Power, Area) studies to inform architectural trade-offs.
• Collaborate with RTL, Compiler, and Runtime teams to instrument systems and correlate model predictions with silicon and prototype results.
• Translate deep learning workload behavior into actionable architectural requirements and measurable design targets.
• Validate that architectural changes produce measurable gains across real-world AI workloads.

Requirements

Core qualifications and skills needed for success. Must-have items are listed first, followed by desirable skills.

• Must-have: Proficiency in C++ and Python for simulation, modeling, and performance analysis.
• Must-have: Experience building or using performance models, simulators, and benchmark suites for heterogeneous compute systems.
• Must-have: Practical experience analyzing single- and multi-node AI workloads and identifying system-level performance bottlenecks.
• Must-have: Experience conducting performance and PPA trade-off studies to guide hardware-software co-design.
• Must-have: Ability to work cross-functionally with RTL, compiler, and runtime teams to instrument and validate performance hypotheses.
• Nice-to-have: Familiarity with RISC-V architectures, custom AI accelerators, or prior work on distributed/multi-chip AI systems.
• Nice-to-have: Background in performance engineering for large-scale or distributed deep learning workloads.

Education Requirements

Not specified.

About the Company

Company: Tenstorrent

Headquarters: Austin, Texas, United States

Tenstorrent is a technology company focused on designing innovative computing solutions. They are known for their expertise in the development of advanced hardware, including ASICs and SoCs, aimed at enhancing performance and efficiency in various applications.