High-Performance Networking Engineer - Supercomputing

xAI•Palo Alto, CA

10d

About The Position

High-Performance Networking Engineer on xAI’s Supercomputing team, you will design and optimize low-latency, high-bandwidth networking solutions using NVIDIA’s RDMA-capable technologies to support some of the world’s largest GPU supercomputing clusters. These clusters drive AI training and inference workloads, demanding cutting-edge performance and scalability.

Requirements

Hands-on experience with NVIDIA RDMA technologies (e.g., GPUDirect RDMA, RoCE, InfiniBand) in HPC or AI supercomputing environments.
Proficiency in programming with Rust, C, or C++ for low-level networking and system optimization.
Familiarity with NVIDIA’s networking stack, including Mellanox drivers, libraries (e.g., libibverbs), and tools (e.g., NVPeerMemory).
Experience optimizing distributed systems with MPI, NCCL, or similar frameworks for GPU-accelerated workloads.
Knowledge of Kubernetes networking and integrating RDMA into containerized environments.

Nice To Haves

Background in AI/ML training workflows and their networking demands (e.g., large-scale parameter synchronization).

Responsibilities

Develop and tune RDMA-based communication systems leveraging NVIDIA GPUs and Mellanox NICs (InfiniBand, RoCE) for ultra-fast data transfer between nodes.
Implement and optimize GPUDirect RDMA to enable direct memory access between GPUs and network interfaces, minimizing CPU overhead.
Integrate RDMA solutions with Kubernetes-based workloads, ensuring seamless operation across distributed compute and storage systems.
Collaborate with AI researchers and infrastructure teams to accelerate data pipelines and collective communications using NCCL and MPI.
Troubleshoot and resolve performance bottlenecks in high-throughput, low-latency networking environments.