Principal Distributed Systems Software Engineer, Collectives
Core
Define and deliver industry-leading AI pre-training and distributed inference performance on AMD GPUs by optimizing collective communication libraries and networking software for multi-GPU and multi-node systems.
Role type
Principal distributed systems software engineer (collective communication & network performance)
Builds
Collective communication libraries (e.g., RCCL, NCCL, MPI), scale-up/scale-out networking software, and performance tooling for LLMs and generative AI models.
Domain
High-performance computing, distributed AI training/inference, GPU networking
Deliverable
production ML models
Required skills
Collective communication library design (RCCL, NCCL, MPI), C++, Linux, network/NIC/GPU hardware architecture, performance modeling, distributed training/inference optimization, open-source collaboration (PyTorch, JAX, vLLM, SGLang), technical leadership
Preferred skills
Deep learning framework integration, heterogeneous compute environments, performance tuning/debugging, mentoring
Technologies
C++, Python, Linux, RDMA, ROCm, PyTorch, JAX, vLLM, SGLang, MPI, NCCL, RCCL
Responsibilities
Define roadmap for collectives and network optimizations; Architect collective communication algorithms; Optimize distributed training/inference for large-scale AI models; Participate in hardware-software co-design; Drive performance modeling frameworks; Collaborate with GPU library, compiler, and runtime teams; Provide guidance on network load balancing and model sharding; Present performance analysis to leadership; Mentor junior engineers
Seniority
Principal, technical leadership & mentorship
