CUDA Engineer - Kernel Optimization

About The Job Mercor connects elite creative and technical talent with leading AI research labs. Headquartered in San Francisco, our investors include Benchmark , General Catalyst , Peter Thiel , Adam D'Angelo , Larry Summers , and Jack Dorsey .

Position: CUDA Engineering Expert

Type: Contract Compensation: $300 per task Location: Remote Role Responsibilities

• Analyze and optimize GPU kernels for performance, efficiency, and hardware utilization.
• Use profiler metrics like L2 cache hit rate and occupancy to guide kernel improvements.
• Review GPU kernel implementations and identify bottlenecks without deep algorithmic background.
• Write and modify C++17, Python, and GPU programming code.
• Apply expertise in CUDA, HIP, and shader programming to improve performance.
• Document optimization decisions clearly, focusing on profiler metrics' utility.

Qualifications Must-Have

• Available to work at least 20 hrs/wk.
• Fluent in core C++ features through C++17.
• Working knowledge of Python and Git.
• Fluent in one GPU programming model like CUDA or HIP.
• 1+ year of professional or research experience with GPUs.
• Strong understanding of GPU profiler performance metrics.

Preferred

• Experience with CUDA, HIP, and CUDA C++ Core Libraries.
• Experience optimizing kernels for NVIDIA Blackwell hardware.
• Familiarity with NSight Compute.
• Prior experience with NVIDIA, AMD, or Qualcomm.
• Open-source contributions related to GPU kernel optimization.

Application Process (Takes 20–30 mins to complete)

• Submit your resume or relevant technical background.
• Qualified applicants may complete a brief technical assessment or submit additional information.

Resources & Support

• For details about the interview process and platform information, please check: https://talent.docs.mercor.com/welcome
• For any help or support, reach out to: [email protected]

PS: Our team reviews applications daily. Please complete your AI interview and application steps to be considered for this opportunity.