Principal Software Engineer, E2E Performance and Goodput — CSP Engagements
Core
Principal Engineer acting as technical focal point for end-to-end performance and goodput for key CSP/hyperscale customers on NVIDIA platforms.
Role type
Principal Software Engineer (CSP Performance Engineering)
Builds
Shared understanding of platform performance, optimization priorities, and validation tooling for CSP workloads.
Domain
High-Performance Computing (HPC) / AI Infrastructure / GPU Systems
Deliverable
production ML models
Required skills
GPU workload profiling, distributed training performance dynamics, statistical methods for performance analysis, full software stack impact analysis, data analysis and visualization, cross-team influence
Preferred skills
Profiling and optimizing distributed training at 1000+ GPU scale, ML infrastructure performance at CSP/hyperscaler, NVIDIA platform familiarity, automated performance regression detection systems, inference workload performance dynamics
Technologies
nsight systems, nsight compute, DCGM metrics, Python, pandas, NVIDIA DGX, NVIDIA HGX, NVLink topology, Megatron-LM, DeepSpeed, FSDP, vLLM, TensorRT-LLM, SGLang
Responsibilities
Drive performance characterization work streams with CSP engineering teams, gather and synthesize CSP performance feedback, update and validate open-source performance tools, ensure CSP tooling reflects latest GPU capabilities, conduct cross-CSP performance comparison and pattern analysis, define test strategies and tooling requirements for performance validation
Seniority
Principal, strategy & mentorship
