CUDA / GPU Performance Engineer (Kernel Optimization)

Gramian Consultancy is a boutique consultancy specializing in IT professional services and engineering talent solutions. It combines software engineering, technical recruiting, and business expertise to provide recruiting, staff augmentation, and outsourcing services.

• Strong professional experience with CUDA programming and GPU kernel optimization.
• Advanced proficiency in C++, ideally in high-performance or systems programming environments.
• Proven experience profiling and tuning GPU workloads for performance.
• Hands-on experience with GPU profiling tools such as NVIDIA Nsight or comparable tools.
• Strong understanding of GPU architecture, memory hierarchy, parallel execution, and synchronization.
• Experience analyzing performance across different GPU hardware generations.
• Hands-on experience with GLSL and/or WebGPU for shader or compute development.
• Ability to document performance findings and technical decisions clearly in English.

• Analyze and optimize CUDA kernels for throughput, latency, and hardware utilization.
• Profile GPU workloads to identify compute, memory, synchronization, and execution bottlenecks.
• Develop and implement targeted kernel optimization strategies.
• Refactor C++ and CUDA codebases for performance, maintainability, and portability.
• Evaluate kernel behavior across different GPU architectures and hardware generations.
• Develop or adapt shader and compute workflows using GLSL and WebGPU.
• Use GPU profiling tools to validate improvements and compare performance.
• Document optimization approaches, benchmarks, findings, and performance gains.
• Contribute technical input to GPU architecture and performance-design discussions.
• Evaluate emerging GPU programming techniques and apply relevant improvements.

Contract: Freelance contractor, paid per completed task. Commitment: Flexible, based on available tasks and project demand. Locations: Fully remote - GLOBAL. Process: Application review, technical assessment, and onboarding.