Senior AI Performance Engineer (Data Center Scale)

Graphcore Early Careers; Milpitas (CA); On-site; USD 180,000 - 240,000; Full time

Job summary

Graphcore is seeking a Staff AI Performance Engineer in Milpitas, California to lead performance work that turns large-scale AI workloads into faster, more efficient systems. You will analyze and optimize training and inference workloads across distributed systems, connecting compute, memory, networking and software behavior.

You will build benchmarks, profile demanding workloads and develop C++ and Python tools for analysis and optimization.

Qualifications

• Strong experience profiling and optimizing AI, ML or HPC workloads.
• Experience with distributed systems and communication libraries such as MPI, NCCL, UCX or libfabric.
• Strong C++ and Python skills with reliable, performance-sensitive tooling.
• Deep understanding of compute, memory and communication behavior in large-scale systems.
• Ability to lead technically complex work and coordinate improvements across teams.
• Familiarity with MLPerf, accelerated architectures, ML frameworks or high-performance interconnects.

Responsibilities

• Lead performance workstreams across hardware and software to optimize AI training and inference on distributed systems.
• Build benchmarks, profile demanding workloads and develop C++/Python tooling for analysis and optimization.
• Translate performance evidence into engineering decisions to improve efficiency, scalability and reliability in data-center scale environments.

Skills

Performance profiling

Distributed systems

C++

Python

System understanding

Interconnects/MLPerf

Tools

MPI

NCCL

UCX

libfabric

Job description

Graphcore is seeking a Staff AI Performance Engineer in Milpitas, California to lead performance work that turns large-scale AI workloads into faster, more efficient systems. You will analyze and optimize training and inference workloads across distributed systems, connecting compute, memory, networking and software behavior.

You will build benchmarks, profile demanding workloads and develop C++ and Python tools for analysis and optimization.