Principal Software Engineer, E2E Performance and Goodput — CSP Engagements

NVIDIA•Austin, CA

About The Position

We're looking for a Principal Engineer to join our CSP Engagements team as the technical focal point for end-to-end performance, working directly with engineering teams of key CSP/hyperscale customers to ensure they achieve various performance targets on NVIDIA platforms. In this role, you will augment NVIDIA's performance and benchmark teams with a dedicated CSP-facing focus. You will drive work streams with CSP engineering teams to build shared understanding of platform performance characteristics, gather and incorporate their workload-specific feedback into NVIDIA's optimization priorities, and validate that performance targets are met in customer-representative configurations. Your cross-CSP visibility enables you to identify patterns and drive systemic improvements in documentation, configuration guidance, and tooling.

Requirements

15+ years of experience in systems performance engineering, ideally in GPU/HPC/ML infrastructure.
BS or MS in Computer Science, Computer Engineering, or related field (or equivalent experience)
Proficiency in GPU workload profiling: nsight systems, nsight compute, DCGM metrics, or equivalent instrumentation
Understanding of distributed training performance dynamics: computation/communication overlap, pipeline bubbles, memory bandwidth utilization, collective efficiency
Statistical methods for performance analysis: regression detection, confidence intervals, A/B comparison at scale
Understanding of how the full software stack impacts performance: driver overhead, collective algorithm selection, memory allocation, scheduling, firmware power management
Strong data analysis and visualization skills (Python, pandas, dashboards).
Customer obsession — genuine passion for understanding why customers aren't achieving expected performance and driving solutions
Ability to communicate performance findings to both deep technical audiences and executive leadership
Demonstrated success influencing multiple engineering teams to prioritize performance improvements

Nice To Haves

Experience profiling and optimizing distributed training at 1000+ GPU scale (Megatron-LM, DeepSpeed, FSDP)
Background in ML infrastructure performance at a CSP/hyperscaler
Familiarity with NVIDIA platforms (DGX, HGX, NVLink topology) and profiling tools
Experience building automated performance regression detection systems for production environments
Understanding of inference workload performance dynamics (vLLM, TensorRT-LLM, SGLang, continuous batching)

Responsibilities

Drive performance characterization work streams with engineering teams of key CSP/hyperscale customers — ensuring they understand platform performance expectations, profiling methodology, and tuning options for their specific workloads
Gather and synthesize CSP performance feedback — identify gaps between expected and actual throughput, and champion optimization priorities back into NVIDIA's CUDA, NCCL, driver, and firmware teams
Ensure key open-source performance and stress tools (e.g., STREAM, GPU Burn, GPU BLAST) are updated and validated for the latest NVIDIA rack-scale systems, GPU architectures, and CPU platforms — so customers and internal teams have reliable baseline measurements from day one
Work closely with CSPs to ensure their own performance and validation tooling reflects the latest GPU capabilities, memory hierarchy changes, and platform-specific tuning parameters
Conduct cross-CSP performance comparison and pattern analysis — identify configuration, software, or workload differences that explain performance gaps between deployments
Collaborate with CSPs to ensure performance-related integration work (profiling infrastructure, benchmark harnesses, config validation) is ready ahead of deployment milestones
Define test strategies and tooling requirements for performance validation — both for NVIDIA internal certification and customer acceptance