GPU Kernel Engineer
San FranciscoFull-timePosted 3d ago
About the role
Relace is building a globally distributed inference cloud. The role focuses on low-level GPU performance, writing and tuning code that runs directly on the GPU, optimizing for faster and cheaper operation, and identifying idle GPU opportunities. Open to strong low-level systems engineers even without CUDA experience.
Responsibilities
- Write and tune code that runs directly on the GPU at the bottom of the stack
- Increase work per second and per dollar of hardware
- Shrink models to reduce running cost without degrading performance
- Identify and fix reasons GPUs sit idle
- Measure performance rigorously (profile before/after) and rely on solid benchmarking
Qualifications
- Five or more years of low-level performance work where system speed mattered (e.g., operating systems, drivers, databases, compilers, trading systems, inference engines)
- Strong C or C++ with low-level proficiency (memory, concurrency, caches, hardware behavior)
- Track record of making something significantly faster with clear explanation of slow points and changes
- Habit of measuring and validating with solid benchmarks
Skills
Required:Low-level performance optimizationC/C++Performance measurement
Benefits
- Competitive compensation plus equity
- On-site SF, five days a week in-person collaboration with whiteboard problem solving
Last updated Oct 22, 2026. Interested? Apply on the company site.
About Relace
Models and infra for coding agents
Relace makes it easy to deploy production-ready coding agents. Our models are co-optimized with infrastructure to achieve SoTA performance: 10k+ token/s code merging and retrieval across million-line repositories in seconds.
More open roles at Relace
- Distributed Systems Engineer3d agoSan Francisco · Infrastructure
- GTM Engineer11mo agoHybrid (San Francisco, CA)
- Machine Learning Scientist11mo agoHybrid (San Francisco, CA)