Senior GPU Networking Architect

Start free →

Or apply yourself on the official page →

About the role

NVIDIA has been transforming computer graphics, PC gaming, and accelerated computing for more than 25 years. It’s a unique legacy of innovation that’s fueled by great technology—and amazing people. Today, we’re tapping into the unlimited potential of AI to define the next era of computing. An era in which our GPU acts as the brains of computers, robots, and self-driving cars that can understand the world. Doing what’s never been done before takes vision, innovation, and the world’s best talent. As an NVIDIAN, you’ll be immersed in a diverse, supportive environment where everyone is inspired to do their best work. Come join the team and see how you can make a lasting impact on the world. We are looking for a Senior GPU Networking Architect to join our networking software group, bringing strong GPU architecture and programming skills to build and improve GPU communication kernels. This role links GPU computing with networking by making sure communication primitives are carefully developed alongside GPU hardware capabilities. Join our team of engineers developing the software foundation for the largest AI systems globally. What you will be doing: • Build, implement, and optimize GPU communication kernels that underpin collective and point-to-point operations in large-scale AI systems. • Leverage deep knowledge of GPU architecture—thread scheduling, memory hierarchy, execution pipelines—to improve kernel efficiency, minimize latency, and overlap computation with communication. • Develop GPU-resident communication primitives and device-side APIs that enable fine-grained, kernel-initiated data movement across nodes and accelerators. • Profile and tune GPU kernels end-to-end, identifying bottlenecks at the intersection of compute, memory, and network, and driving targeted optimizations. • Collaborate with network software, hardware, and AI framework teams to co-design communication strategies that align with GPU execution patterns and emerging model architectures. • Build proofs-of-concept, conduct experiments, and perform quantitative modeling to evaluate and validate new communication strategies before committing them to production. • Contribute to the evolution of programming models that expose GPU-aware networking capabilities to application developers. What we need to see: • 5+ years of hands-on CUDA programming, including writing and optimizing non-trivial GPU kernels. • M.Sc. or equivalent experience in computer science, computer engineering, or a closely related field. • Strong understanding of GPU architecture fundamentals: warp scheduling, shared memory, L2 cache, memory coalescing, occupancy tuning, and asynchronous execution. • Experience with systems-level C/C++ development in performance-critical environments. • Familiarity with GPU data movement mechanisms such as GPUDirect RDMA and GPU-initiated communication. • Ability to read and reason about GPU performance profiles (e.g., Nsight Compute, Nsight Systems) and translate observations into actionable optimizations. • Strong collaboration skills in a multi-national, interdisciplinary environment. Ways to stand out from the crowd: • Experience developing or optimizing communication kernels in libraries such as NCCL, NVSHMEM, or similar GPU-aware communication frameworks. • Understanding of distributed deep learning parallelism techniques, including data parallelism, tensor parallelism, pipeline parallelism, expert parallelism, and mixture-of-experts parallelism, and the communication patterns they impose on GPU kernels. • Background in RDMA, InfiniBand, high-speed networking, and GPU system topology, including NVLink, NVSwitch, PCIe, and network fabrics, and their impact on communication kernel design. • Experience with overlap techniques such as kernel

View the official posting →