Cluster Architect

Andromeda is a platform for buying, selling, and operating compute. The Cluster Architect owns reference designs and turns customer requirements into complete GPU cluster designs, serving as the technical authority for each build.

Responsibilities:

• Reference designs: Standard cluster designs for current and next-generation NVIDIA platforms
• Customer designs: Configurations, BOMs, and rack layouts built from each customer's workload, scale, and site
• Workload analysis: Analysis of customer training and inference workloads to set design tradeoffs across latency, bandwidth, and GPU density
• Power-constrained design: Designs that fit fixed site power budgets, maximizing GPU density while reserving headroom for storage, networking, and services
• Quote and BOM review: Technical review of vendor quotes and BOMs
• Acceptance criteria: Burn-in, performance, and acceptance tests that define when a cluster is ready for service
• Site fit: Power, cooling, and floor requirements, validated against each candidate data center