TPU Performance Engineer – ML Hardware & Compiler Co-Design

Google; Sunnyvale (CA); On-site; USD 150,000 - 210,000; Full time

Benefits offered by this job

Bonus target

Equity

Benefits

Job summary

Google Sunnyvale seeks a software engineer to bridge ML workloads with TPU hardware, focusing on performance analysis and optimization of distributed systems and compiler architectures. You will identify bottlenecks, model workloads, and propose architectural improvements with measurable impact.

Responsibilities include developing benchmarking strategies, driving hardware-software co-design for ML accelerators, and collaborating with cross-functional teams to deliver production-ready

Qualifications

• Candidates have a Bachelor's degree and 2+ years in software development, performance analysis, and computer architecture.
• Preferred: Master's or PhD in Computer Science; experience with data structures and algorithms.

Responsibilities

• Bridge ML workloads with TPU hardware through performance analysis and optimization of distributed systems and compiler architectures.
• Develop benchmarking strategies and drive hardware-software co-design to optimize ML accelerator architectures for production models.

Skills

Software Development

Performance Analysis

Computer Architecture

Performance Modeling

Data Structures

Algorithms

Distributed Systems

Compiler Architectures

XLA

Hardware-Software Co-design

Benchmarking

Workload Characterization

Simulator Tools

Hardware Cost-models

ML Infrastructure

Data Analysis

Education

Bachelor's degree

Master's degree or PhD (preferred)

Tools

XLA

Simulator Tools

Hardware Cost-models

Job description

Google Sunnyvale seeks a software engineer to bridge ML workloads with TPU hardware, focusing on performance analysis and optimization of distributed systems and compiler architectures. You will identify bottlenecks, model workloads, and propose architectural improvements with measurable impact.

Responsibilities include developing benchmarking strategies, driving hardware-software co-design for ML accelerators, and collaborating with cross-functional teams to deliver production-ready