Principal Engineer - Perf and Benchmarking
Core
Technical lead for planet-scale performance data ingestion, storage, and analysis, plus driving industry-leading MLPerf benchmarking publications.
Role type
Principal Engineer (Performance & Benchmarking)
Builds
Planet-scale performance data warehouse; MLPerf (Training & Inference) submissions; Kubernetes-native benchmarking service.
Domain
AI Infrastructure / High-Performance Computing / Cloud Data Warehousing
Deliverable
production ML models | dashboards & analysis | infrastructure
Required skills
Distributed systems architecture, HPC/cloud services, GPU performance tuning (CUDA, NCCL, RDMA), Model-server stacks (Triton, vLLM, TensorRT-LLM), Distributed training frameworks (PyTorch FSDP, DeepSpeed, Megatron-LM), Kubernetes, ML control planes, Time-series databases, Log-structured merge trees (LSM), Supply-chain integrity (SBOMs, Cosign)
Preferred skills
MLPerf submission experience, OSS contributions, Multi-region fleet benchmarking, Publications on ML performance
Technologies
Kubernetes, SUNK, Kueue, Kubeflow, Prometheus, Grafana, OpenTelemetry, NVIDIA (Megatron-LM, TensorRT-LLM, DGX cloud), PyTorch, DeepSpeed, vLLM, Triton, KServe, ONNX Runtime
Responsibilities
Define multi-year benchmarking strategy and roadmap; Lead end-to-end MLPerf submissions; Design and maintain Kubernetes-native benchmarking service; Build CI/CD pipelines and observability integrations; Partner with NVIDIA and OSS projects for optimizations.
Seniority
Principal, strategy & mentorship