Senior GPU Networking Architect
Core
Design and optimize GPU communication kernels and device-side APIs to enable high-performance data movement in large-scale AI systems.
Role type
Senior GPU Networking Architect (Systems/Software)
Builds
GPU communication primitives, device-side APIs, and programming models for distributed AI training and inference.
Domain
High-performance computing, GPU architecture, distributed deep learning, networking.
Deliverable
production ML models
Required skills
CUDA programming, GPU architecture (warp scheduling, memory hierarchy), systems-level C/C++, GPU data movement (GPUDirect RDMA), performance profiling (Nsight), distributed deep learning parallelism, RDMA/InfiniBand/NVLink.
Preferred skills
NCCL/NVSHMEM development, kernel pipelining/persistent kernels, large-scale LLM workload optimization (PyTorch, TensorRT-LLM, vLLM), disaggregated serving architectures.
Technologies
CUDA, C/C++, NCCL, NVSHMEM, Nsight Compute, Nsight Systems, InfiniBand, NVLink, NVSwitch, PyTorch, TensorRT-LLM, vLLM.
Responsibilities
Build and optimize GPU communication kernels for collective and point-to-point operations; develop GPU-resident communication primitives and device-side APIs; profile and tune kernels to minimize latency and overlap compute with communication; collaborate with hardware and AI framework teams to co-design communication strategies; build proofs-of-concept and conduct quantitative modeling for new strategies; contribute to programming models exposing GPU-aware networking capabilities.
Seniority
Senior, hands-on IC
