CUDA Kernel Performance Engineer

Mercor; San Francisco (CA); On-site; USD 96,000 - 165,000; Part time

Job summary

Mercor is seeking GPU kernel optimization experts to contribute to a project with a leading AI lab. This contract-based opportunity is designed for freelancers with strong C++ skills, practical GPU programming experience, and the ability to squeeze performance out of modern GPU architectures.

You will analyze, optimize, and reason about GPU kernels across modern hardware, using profiler-guided analysis. Expect to write C++17 and Python code, apply CUDA or HIP, and document decisions clearly.

Qualifications

• Available to work at least 20 hrs/wk.
• Proficient in C++ through C++17.
• Experience with Python and Git.
• Skilled in GPU kernel programming models (CUDA, HIP, Slang, HLSL, GLSL).
• At least 1 year of GPU experience.
• Strong understanding of GPU profiler metrics.
• Ability to optimize kernels with minimal context.

Responsibilities

• Analyze and optimize GPU kernels for performance and hardware utilization.
• Use profiler metrics (L2 cache, occupancy) to guide improvements.
• Review kernel implementations and identify bottlenecks.
• Write and modify C++17, Python, and GPU code.
• Apply CUDA, HIP, shader programming to improve performance.
• Document optimization decisions clearly.

Skills

C++17

Python

GPU programming

CUDA

HIP

Kernel optimization

Profiling tools

Git

Shader programming

Tools

NSight Compute

CUDA

HIP

PTX

Job description

Mercor is seeking GPU kernel optimization experts to contribute to a project with a leading AI lab. This contract-based opportunity is designed for freelancers with strong C++ skills, practical GPU programming experience, and the ability to squeeze performance out of modern GPU architectures.

You will analyze, optimize, and reason about GPU kernels across modern hardware, using profiler-guided analysis. Expect to write C++17 and Python code, apply CUDA or HIP, and document decisions clearly.