CUDA Engineer - Kernel Optimization

Obsidian; Berlin; Vor Ort; EUR 83.000 - 165.000; Teilzeit

Zusammenfassung

Mercor seeks GPU kernel optimization experts to contribute to a cutting-edge AI-lab project. This contract-based opportunity targets freelancers with strong C++17 and GPU programming skills, capable of improving kernel performance across modern hardware.

You will analyze and optimize GPU kernels, guided by profiler metrics, and document your optimization decisions clearly for reproducibility. 20 hours per week minimum, based in Berlin.

Qualifikationen

• 1+ year of professional or graduate-level GPU experience.
• Strong C++17 and Python programming skills.
• Experience with CUDA or HIP and kernel programming.
• Knowledge of GPU profiler metrics and performance tuning.
• Familiarity with NSight Compute is a plus.
• Able to work at least 20 hours per week.

Aufgaben

• Analyze and optimize GPU kernels for performance, efficiency, and hardware utilization
• Use profiler metrics such as L2 cache hit rate, L2 throughput, occupancy, and related signals to guide kernel improvements
• Review GPU kernel implementations and identify bottlenecks without requiring extensive background in the underlying algorithms
• Write, modify, and reason about C++17, Python, and GPU programming code
• Apply CUDA, HIP, shader programming, or related kernel programming expertise to improve performance outcomes
• Document optimization decisions clearly, including when specific profiler metrics are or are not useful

Kenntnisse

C++17

Python

GPU programming

Kernel optimization

Profiling

CUDA

HIP

Slang

HLSL

GLSL

Inline PTX assembly

Tensor core optimization

Tools

NSight Compute

CUDA Core Libraries

GPU hardware know-how

Jobbeschreibung

1. Role Overview

Mercor is seeking GPU kernel optimization experts to contribute to a project with a leading AI lab. This opportunity is designed for freelancers with strong C++ skills, practical GPU programming experience, and the ability to improve kernel performance using profiler-guided analysis. You'll help evaluate, optimize, and reason about GPU kernels across modern hardware environments. This is a contract-based opportunity for specialists who enjoy squeezing performance out of modern GPU architectures.

2. Key Responsibilities

• Analyze and optimize GPU kernels for performance, efficiency, and hardware utilization

• Use profiler metrics such as L2 cache hit rate, L2 throughput, occupancy, and related signals to guide kernel improvements

• Review GPU kernel implementations and identify bottlenecks without requiring extensive background in the underlying algorithms

• Write, modify, and reason about C++17, Python, and GPU programming code

• Apply CUDA, HIP, shader programming, or related kernel programming expertise to improve performance outcomes

• Document optimization decisions clearly, including when specific profiler metrics are or are not useful

3. Ideal Qualifications

• Available to work at least 20 hrs/wk

• Fluent in core C++ features through C++17

• Working knowledge of Python and Git

• Fluent in at least one GPU programming model, such as CUDA, HIP, Slang, HLSL, GLSL, or related kernel programming

• At least 1 year of professional or graduate-level research experience working with GPUs

• Strong understanding of GPU profiler performance metrics and how to use them to optimize kernels

• Ability to optimize GPU kernels without needing deep prior context on every algorithm

• Experience with CUDA, HIP, CUDA C++ Core Libraries, inline PTX assembly, or tensor core-level optimization is a plus

• Experience optimizing kernels for NVIDIA Blackwell hardware is a plus

• Familiarity with NSight Compute is a plus

• Prior experience with GPU hardware organizations such as NVIDIA, AMD, or Qualcomm is a plus

• Open-source contributions related to GPU kernel optimization are a plus