Senior Staff Software Engineer, ML Performance, GPU
Overview:
In this role you work on optimizing GPU performance for Gemini models and LLMs at scale. You will collaborate with DeepMind and XLA to push toward state-of-the-art efficiency, including custom kernels and compiler optimizations. You'll run architecture-level simulations, perform roofline analyses, and build tooling to track performance across Google's fleets. You will impact how models train and serve on AI accelerators, shaping scalable, high-performance ML infrastructure.
Responsibilities:
- Identify performance opportunities and drive XLA:GPU/Triton optimizations toward peak efficiency
- Collaborate with DeepMind on onboarding new LLMs and optimizing deployment and training on GPUs at very large scale
- Perform architecture-level GPU simulations and roofline analyses to guide internal teams
- Analyze performance metrics to identify bottlenecks and implement scalable solutions fleet-wide
- Partner with tooling and fleet metrics teams to build observability and extract workload metrics
- Develop and promote compiler and kernel-level optimizations (sharding, quantization, custom kernels)
Key requirements:
- Bachelor's degree or equivalent practical experience 8 years of software development experience 7 years leading technical project strategy, ML design, and end-to-end ML optimization
- Experience with modern GPU architectures, memory hierarchies, and performance bottlenecks
- Experience with low-level GPU programming (CUDA, Triton, CUTLASS, etc.) and performance engineering
- Experience with modern LLMs and deployment on AI acceleratorsleadership and cross-functional collaborationproblem-solving mindset with a bias for resultseffective communication of complex technical conceptsCUDATritonCUTLASS