AI Kernel Performance Engineering Manager

Meta; Bellevue, Menlo Park, New York (WA, CA, NY); On-site; USD 230,000 - 320,000; Full time

Job summary

Meta is seeking a Software Engineering Manager for the MTIA Software team within the AI & Compute Foundation. You will lead kernel and performance engineers, owning roofline-level performance on MTIA accelerators and guiding technical strategy across kernel, compiler, runtime, and libraries.

You’ll partner with silicon architecture to co-design features, drive measurable improvements, and mentor engineers to grow technically while delivering high-impact results for billions of users.

Qualifications

• Experience leading software engineers in a system-level or HPC environment.
• Proven ability to set technical direction and develop measurable outcomes.

Responsibilities

• lead a kernel and performance engineering team to achieve roofline-level performance on MTIA accelerators.
• Drive technical strategy for kernel development and optimization across stack layers.
• Collaborate with silicon architecture teams on hardware/software co-design.
• Contribute to architectural decisions while enabling team velocity.

Skills

Kernel optimization

System-level software

Leadership

Roadmap planning

Performance profiling

Distributed systems

HPC & ML systems

PyTorch integration

Hardware/software co-design

C++/Python

Transformer kernels

NCCL/RCCL

Pre-silicon bring-up

Compiler/toolchain

Quantization/low precision

Safety and ethics in AI

Tools

CUTLASS

cuBLAS

cuDNN

Triton

oneDNN

Job description

Meta is seeking a Software Engineering Manager for the MTIA Software team within the AI & Compute Foundation. You will lead kernel and performance engineers, owning roofline-level performance on MTIA accelerators and guiding technical strategy across kernel, compiler, runtime, and libraries.

You’ll partner with silicon architecture to co-design features, drive measurable improvements, and mentor engineers to grow technically while delivering high-impact results for billions of users.