Distributed Systems Software Engineer, Collectives

THE ROLE: As a member of the Collectives and Network Performance team, you will contribute to developing, testing and optimizing collective communication and networking software for AMD GPUs. You will help improve training and inference performance across multi-GPU and multi-node systems, working with senior engineers and with internal GPU library, runtime and ROCm teams, and learning how to integrate optimizations across AMD GPU generations and a range of AI models.

THE PERSON: Seeking a motivated C++ developer with a solid foundation in Linux environments, computer architecture and networking concepts. The ideal candidate completes well-defined tasks with guidance, learns quickly, and collaborates effectively within a high-performing team. You are comfortable profiling and debugging performance issues and are interested in how AI model architectures map onto software and hardware. Familiarity with GPU programming and distributed training or inference is a plus.

KEY RESPONSIBILITIES:

• Develop & Test Features: Implement, test and debug features in collective communication libraries and network software, under the guidance of senior engineers.
• Profile & Analyze Workloads: Profile LLM, diffusion, multimodal, RecSys and generative AI workloads on single-node and multi-node systems, and help identify performance bottlenecks.
• Run Benchmarks & Experiments: Run performance experiments across AMD GPU generations and network configurations, and summarize results for the team.
• Contribute to Tooling: Contribute to tools and infrastructure for performance measurement, regression tracking and reporting.
• Support Model & Framework Enablement: Support new model enablement and out-of-box performance efforts across AI frameworks, and work with internal teams and open-source communities to get changes reviewed and integrated.
• Software Engineering Best Practices: Follow sound engineering practices, including code reviews, testing and documentation, to deliver robust, maintainable code.
• Learn & Grow: Build expertise in collectives, networking and GPU performance through mentorship, code reviews, knowledge sharing and hands-on project work.

…