Software Engineer, AI Kernels & Performance Optimization - MTIA Software

Job Title

Software Engineer, AI Kernels & Performance Optimization - MTIA Software

Role Summary

Ownership of high-performance kernel development and performance optimization for Meta's MTIA AI accelerators. The role sits on the MTIA Software team within AI & Compute Foundation and focuses on kernel implementation, numerics, tooling, and hardware/software co-design from pre-silicon through production.

This is a hands-on engineering role responsible for delivering roofline-level kernels, building kernel authoring frameworks, and driving cross-team performance improvements for production models.

Experience Level

Mid-level (experienced engineer). The posting requests approximately 6+ years of relevant professional experience and expects senior-level impact on kernel performance and architecture interaction.

Responsibilities

Deliver production-quality kernels and lead performance work across stack and silicon/product boundaries.

• Design, implement, and optimize compute and communication kernels (GEMM, attention variants, normalization, elementwise/reduction fusions, sparse/quantized paths) for MTIA accelerators.
• Profile and root-cause performance across instruction scheduling, memory hierarchy, DMA, on-chip interconnect, and multi-device collectives; drive fixes to kernel, compiler, runtime, or hardware as appropriate.
• Build and extend kernel authoring frameworks, templates, and libraries to broaden high-performance coverage and lower the expertise barrier.
• Deliver and maintain PyTorch operator coverage for recommendation, ranking, and generative AI workloads in both eager and compiled paths.
• Partner with silicon architecture and design teams to quantify feature value, characterize rooflines pre-silicon, and influence hardware decisions.
• Develop numerics and low-precision solutions and software mitigations for hardware limitations.
• Work with compiler, runtime, and framework teams to produce end-to-end model performance wins, not just microbenchmark improvements.
• Set technical direction for kernel domains, write design documents, and mentor other engineers on performance methodology.

Requirements

Must-have skills and experience for the role.

• 6+ years of professional experience in high-performance computing, accelerator kernel development, compiler backends, or systems performance engineering.
• Proficiency in C++ and Python, including low-level systems programming, templates/generic programming, and performance-critical code.
• Experience writing and optimizing kernels for parallel architectures (GPU CUDA/ROCm/HIP, SYCL/OpenCL, TPU/AI ASICs, or SIMD/vector CPU targets).
• Working knowledge of computer architecture relevant to performance: memory hierarchies, bandwidth/latency, occupancy, scheduling, vectorization, and synchronization.
• Measurement-driven approach to performance: ability to build roofline or analytical models, profile against them, and explain residual gaps.
• Experience with pre-silicon development, simulators/emulators, or first-light bring-up is desirable but not mandatory for all candidates.

Education Requirements

Bachelor's degree in Computer Science, Computer Engineering, or a related technical field, or equivalent practical experience. The posting also references advanced degrees as relevant in preferred qualifications and explicitly allows "equivalent practical experience." No specific certifications were required.

About the Company

Company: Meta Platforms

Headquarters: Menlo Park, California, United States

American technology company that develops social networking products (Facebook, Instagram, WhatsApp) and invests in virtual/augmented reality hardware and software through Reality Labs, focusing on connectivity, advertising, and immersive computing experiences.