General Motors
Senior ML Accelerator Engineer - GPU
$170K–$258K
Sunnyvale, California, United States of America +4 · Hybrid · Senior · Full-time · Posted Oct 26, 2026
- Location
- Sunnyvale, California, United States of America
Remote - Washington
Austin, Texas, United States of America
San Francisco, California, United States of America
Warren, Michigan, United States of America - Experience
- Minimum 2+ years of relevant industry experience or equivalent experience; preferred qualifications also state 2+ years of relevant industry experience or equivalent experience.
- Education
- BS, MS or PhD in CS, or related technical field.
About the job
Develop and optimize GPU kernels, custom operators, and supporting tools for on-vehicle machine-learning inference in ADAS and autonomous driving.
What you'll do
- Design, benchmark, and optimize CUDA kernels and custom operators for vehicle inference workloads.
- Build tools and infrastructure to profile, debug, and validate kernels across the AV software stack.
- Collaborate with compiler, runtime, architecture, and deployment teams to deliver reliable production libraries.
What they ask for
- BS, MS, or PhD in CS or a related technical field.
- Excellent CUDA and GPU programming skills, including knowledge of parallel programming patterns and GPU architecture.
- Hands-on experience benchmarking, profiling, debugging, and optimizing accelerator kernels or libraries using the NSight suite or similar tools.
- Strong C++ programming skills and background in software architecture, library design, and system performance.
Nice to have
- Experience with tensor core programming, CUTLASS, and/or CuTe.
- Experience with transformer-based ML models, low-latency or real-time systems, or lower-level accelerator software.
Pay
$170,100 to $258,300 USD base salary
Tools
CUDA, NSight, C++, CUTLASS, CuTe
AI focus
Enables efficient on-vehicle ML inference for ADAS and autonomous driving by developing GPU kernels, custom operators, and performance tooling.
Model-written summary · the original posting is authoritative