GPU Kernel Engineer
San Francisco • FullTime
Posted 13h ago
Job Location
San Francisco
Tech Stack
Remote Work Policy
On-site
Employment Type
FullTime
Categories
AI Infrastructure Engineer
About the job
Relace is building a globally distributed inference cloud, competing on price, speed, and reliability. We process over 800 billion tokens daily and have experienced significant growth, backed by a $23 million Series A funding round. Our team is small, offering substantial responsibility and growth opportunities. We develop our own stack, running on Kubernetes across multiple clouds, where performance directly impacts business success. This role focuses on the lowest level of our stack, optimizing code that runs directly on GPUs to maximize work per second and dollar, shrink models for cost-efficiency without sacrificing quality, and eliminate GPU idle time.
Responsibilities
- Write and tune code that runs directly on the GPU.
- Optimize GPU performance to increase work per second and per dollar.
- Shrink models to reduce running costs without compromising quality.
- Identify and resolve reasons for GPU idleness.
Requirements
- Five or more years of experience in low-level performance optimization (e.g., operating systems, drivers, databases, compilers, trading systems, inference engines).
- Strong proficiency in C or C++ with deep understanding of low-level concepts like memory, concurrency, caches, and hardware behavior.
- Demonstrated ability to significantly improve system speed, with clear explanations of performance bottlenecks and solutions.
- A habit of rigorous measurement, profiling, and suspicion of unreliable test setups.
Benefits
- Competitive compensation
- Equity