AI Cluster Architect — Power-Aware GPU Scaling

Webhosting; Northern (KY); Hybrid; USD 165,000 - 185,000; Full time

Benefits offered by this job

Medical Benefits

401(k) Match

PD Reimbursement

PTO & Holidays

Sabbatical

Remote setup stipend

Internet Reimbursement

Gym Reimbursement

Job summary

Vultr seeks an AI Cluster Architect to design and refine large-scale GPU clusters within fixed power envelopes. You will optimize GPU density while accounting for compute, storage, networking, cooling, and facility constraints, balancing GPU power, fabric performance, and service density.

You will evaluate InfiniBand, RoCE, SpectrumX designs, model power usage, and develop templates for scalable deployments across sites, collaborating with vendors to enable 100k+ GPU deployments.

Qualifications

• Experience designing large-scale GPU clusters within power constraints.
• Familiarity with InfiniBand, RoCE, SpectrumX architectures.

Responsibilities

• Architect large-scale GPU clusters within fixed site power budgets.
• Model and validate power consumption across the cluster BOM.
• Evaluate fabric networking architectures (InfiniBand, RoCE, SpectrumX) and multi-plane/topology options.
• Determine network scale limits based on switch radix, link speed, topology, and blocking requirements.
• Gather SKU-level power and thermal specs for GPUs, NICs, switches, DPUs, storage, and servers.
• Develop power-aware cluster templates and capacity-planning models for multiple sites.
• Document architecture decisions and lifecycle considerations for deployment.
• Provide guidance on future-proofing with next-gen GPUs, NICs, or fabrics.
• Collaborate with vendors on novel fabric architectures for 100k+ GPUs

Skills

GPU cluster design

Power-aware design

High-performance computing

Networking fabrics

Job description

Vultr seeks an AI Cluster Architect to design and refine large-scale GPU clusters within fixed power envelopes. You will optimize GPU density while accounting for compute, storage, networking, cooling, and facility constraints, balancing GPU power, fabric performance, and service density.

You will evaluate InfiniBand, RoCE, SpectrumX designs, model power usage, and develop templates for scalable deployments across sites, collaborating with vendors to enable 100k+ GPU deployments.