4 hours ago
Responsibilities
- Write and tune low-level code that runs directly on GPUs.
- Improve hardware throughput and cost efficiency for inference workloads.
- Shrink models without reducing their quality.
- Investigate and resolve causes of GPU idleness.
- Profile systems before and after changes to identify performance bottlenecks.
- Explain performance problems precisely and implement changes that make systems significantly faster.
Requirements
- Around five or more years of low-level performance work where system speed was critical, such as operating systems, drivers, databases, compilers, trading systems, or inference engines.
- Strong C or C++ skills and comfort with memory, concurrency, caches, and hardware behavior.
- Demonstrated success making a system significantly faster and explaining the causes and solutions precisely.
- Strong measurement and profiling practices with careful test setup.
- GPU or CUDA experience is not required.
Benefits
- Competitive compensation plus equity.
- In-person work in San Francisco five days per week.
- Opportunity to work on a small team with substantial responsibility and room to grow.
Tech Stack
Categories
