Principal Kubernetes GPU Infrastructure Engineer

Advanced Micro Devices, Inc•Santa Clara, CA

7d•Hybrid

About The Position

At AMD, our mission is to build great products that accelerate next-generation computing experiences—from AI and data centers, to PCs, gaming and embedded systems. Grounded in a culture of innovation and collaboration, we believe real progress comes from bold ideas, human ingenuity and a shared passion to create something extraordinary. When you join AMD, you’ll discover the real differentiator is our culture. We push the limits of innovation to solve the world’s most important challenges—striving for execution excellence, while being direct, humble, collaborative, and inclusive of diverse perspectives. Join us as we shape the future of AI and beyond. Together, we advance your career. THE ROLE: As a Principal AI Infrastructure Solution Engineer, you will partner with AMD’s AI software teams and customers to enable large‑scale LLM training and inference on AMD Instinct GPUs. You will design and validate production‑ready Kubernetes architectures and translate inference frameworks such as vLLM and SGLang into deployable customer solutions. Your work will accelerate customer time‑to‑production and strengthen AMD’s leadership in AI infrastructure. THE PERSON: You are a solution‑oriented AI infrastructure engineer with strong expertise in GPU‑accelerated computing and large‑scale AI deployments. You excel at translating complex technologies into customer‑ready solutions and delivering production‑grade Kubernetes‑based inference and training systems. You bring hands‑on experience with Kubernetes‑native distributed training, including scheduling, topology‑aware GPU placement, and operating resilient, high‑performance AI workloads at scale.

Requirements

Deployed and operated large‑scale GPU clusters for production AI training and inference
Deep expertise in Kubernetes GPU orchestration (operators, device plugins, scheduling, multi‑tenancy, observability)
Hands‑on experience with distributed training on Kubernetes (Kubeflow, MPI Operator, Volcano, Kueue, Ray)
Strong knowledge of gang scheduling, elastic jobs, quotas, priority, and shared GPU environments
Tuned Kubernetes networking and storage for AI workloads (high‑performance CNI, RDMA where applicable, scalable checkpointing)
Implemented ML observability for training (GPU/comms metrics, step‑time analysis, SLO‑driven ops)
Experience in AI/ML infrastructure, solution architecture, and production GPU deployments
Proven success enabling customers through complex AI platform deployments and migrations
Strong background working across engineering and customer‑facing roles
Understanding of AI accelerator architectures and inference optimization techniques
Experience operationalizing Kubernetes‑based distributed training at scale

Nice To Haves

Open‑source contributions or AI infrastructure community engagement (plus)

Responsibilities

Design and deliver reference architectures for LLM training and inference on AMD GPUs, from single‑node to multi‑datacenter deployments using Kubernetes and SLURM.
Architect and validate Kubernetes‑based distributed training stacks for large‑scale LLM workloads on AMD GPUs.
Define and implement gang scheduling and topology‑aware GPU placement for multi‑node training workloads.
Enable Kubernetes‑native training controllers including Kubeflow Training Operator, MPI Operator, Volcano, and Kueue.
Partner with enterprise customers and cloud providers to deploy and optimize production AMD GPU clusters for distributed inference and multi‑tenant workloads.
Implement and validate GPU orchestration using Kubernetes GPU Operator, device plugins, metrics exporters, and SLURM controllers.
Benchmark and optimize LLM inference frameworks (vLLM, SGLang) on AMD hardware, producing customer‑ready performance playbooks.
Develop repeatable benchmarks for Kubernetes‑based distributed training, covering scaling efficiency, step time, communication, and checkpointing.
Create tuning guides for RCCL/NCCL‑equivalent communication, CPU/GPU affinity, interconnect utilization, and workload‑specific optimizations.
Serve as the feedback loop between customers and AMD engineering, translating requirements into validated performance improvements.