AI Benchmarking & Performance Architect

CoreWeave; Sunnyvale (CA); On-site; USD 206,000 - 333,000; Full time

Benefits offered by this job

Medical, dental, and vision insurance

Equity awards

401(k) with match

Tuition reimbursement

Paid parental leave

Job summary

CoreWeave, a leader in AI infrastructure, seeks a Principal Engineer to head the Benchmarking & Performance team. You will own the data warehouse for performance events across global data centers and drive industry-leading benchmarking publications with MLPerf references and collaboration with NVIDIA and OSS communities.

You will define strategy, build teams, ensure reproducible methodologies, and coordinate with software and hardware partners to deliver credible, published results while shaping

Qualifications

• 10+ years building distributed systems or HPC/cloud services
• Proven track record of architecting planet-scale data systems
• Deep understanding of GPU performance and model-server stacks
• Proficient with Kubernetes and ML control planes

Responsibilities

• Define the multi-year benchmarking strategy and roadmap for models/workloads
• Lead end-to-end MLPerf Inference and Training submissions with runbooks and audits
• Design a Kubernetes-native benchmarking service across SUNK, Kueue, Kubeflow pipelines
• Build CI/CD pipelines and observability integration for scalable benchmarks
• Partner with NVIDIA, OSS projects, and Sales/SEs to provide data for evaluations

Skills

Distributed systems

HPC/Cloud services

GPU performance

Kubernetes

Communication with executives

Tools

CUDA

cuDNN

TensorRT

NCCL

Kubeflow

Job description

CoreWeave, a leader in AI infrastructure, seeks a Principal Engineer to head the Benchmarking & Performance team. You will own the data warehouse for performance events across global data centers and drive industry-leading benchmarking publications with MLPerf references and collaboration with NVIDIA and OSS communities.

You will define strategy, build teams, ensure reproducible methodologies, and coordinate with software and hardware partners to deliver credible, published results while shaping