GPU Engineer / Senior C++ CUDA Developer

🔎 Seeking GPU Engineer / Senior C++ CUDA Developer

Developing a GPU-native data engine core for AI and analytics. We are looking for an engineer with experience in writing and optimizing custom GPU kernels in industrial systems.

📍 Location: Russian Federation, Republic of Belarus 💻 Format: remote 🕐 Employment: full-time, long-term project 💰 Salary: up to 320,000 RUB per month 📄 Formalization: sole proprietor (IP) only

📌 About the project The engine processes data directly from Parquet and Iceberg in S3-compatible storage. JOIN, aggregations, sorts, and other operations are performed on the GPU. The system supports multiple GPUs and processing of volumes exceeding video memory by offloading intermediate data to NVMe.

The first task is to port the engine from NVIDIA to AMD Instinct while maintaining a single codebase for CUDA and HIP. Subsequently, it involves developing the computational core and optimizing for new generations of accelerators.

📌 What you will do: • Develop and optimize CUDA/HIP kernels: JOIN, sorts, aggregations, string and window functions. • Develop GPU memory management, thread scheduling, and kernel fusion. • Work with GPU compression and porting nvCOMP components to ROCm. • Profile performance, identify bottlenecks, and verify the correctness of results. • Maintain builds for NVIDIA and AMD. • Prepare specifications for AI agents, verify their code, and make architectural decisions.

📌 REQUIREMENTS: • 5+ years of C++ development experience, including C++17, and 2+ years of GPU programming in industrial systems. • Proficient CUDA: custom kernels, memory hierarchy, streams, copy/compute overlap, occupancy, and register pressure. • Understanding of DBMS architecture: SQL, execution plans, JOIN and aggregation algorithms, columnar storage. • Practical experience with AI agents – Claude Code, Cursor, Codex, or equivalents: from task formulation through specification to result verification. • Linux, CMake, Docker. • English for working with code, documentation, and discussing changes.

📌 Will be a plus: Thrust/CUB, HIP/ROCm, Nsight Systems/Compute, rocprof, RAPIDS, multi-GPU, GPU codecs, and code porting between computational platforms.

ROCm experience is not mandatory – deep knowledge of CUDA and the ability to optimize GPU code are more important.

The team uses spec-driven development: AI agents perform a significant portion of mechanical changes, while the engineer is responsible for specifications, architecture, performance, and code verification.

📩 To apply, please write to me via direct message or Telegram: recruiter contact

❗️Send your resume and briefly indicate:

• which CUDA kernels you have developed;
• which optimization tasks you have solved and what results you achieved;
• whether you have experience with HIP/ROCm and development with AI agents;
• when you are ready to start and if cooperation as an IP is possible.