Cluster Architect

Role overview Design the reference architectures and customer-specific configurations for large GPU clusters used in AI training and inference. You’ll translate workload and site requirements into end-to-end designs spanning compute, networking, storage, management, power and cooling, and serve as the technical authority through each build.

Responsibilities - Create standard designs for current and next-generation NVIDIA platforms, and adapt them to each customer’s workload, scale, and data center. - Produce cluster configurations, bills of materials, and rack layouts. - Analyze training and inference needs to balance latency, bandwidth, GPU density, and other design constraints. - Fit designs within site power budgets while accounting for networking, storage, and service capacity. - Review vendor quotes and bills of materials, and define burn-in, performance, and acceptance criteria. - Validate power, cooling, and floor requirements against candidate sites.

Requirements - Experience designing and delivering production GPU clusters with at least 1,000 GPUs. - Detailed knowledge of NVIDIA HGX and NVL72 systems and SuperPOD reference architectures. - Working knowledge of InfiniBand or Spectrum-X fabrics, parallel storage systems, and Slurm or Kubernetes. - Understanding of rack power, liquid cooling, and how facility limits affect cluster design.

Nice to have - Experience with NCCL, HPL, or MLPerf benchmarking and cluster acceptance. - Delivery experience with GB200 or GB300 NVL72 systems. - An NVIDIA architect certification or equivalent.

Benefits and work setup Full-time role based in San Francisco or remote within the United States. Benefits include medical, dental, and vision coverage for employees and dependents, a 401(k), and unlimited paid time off.