Senior AI Inference Performance Engineer — Scale GPUs

NVIDIA; California (MO); On-site; USD 124,000 - 196,000; Full time

Benefits offered by this job

Equity eligibility

Job summary

NVIDIA is seeking a Senior Inference Performance Engineer to push performance limits for large-scale AI inference. You will optimize AI model execution, distill insights into reusable methods, and collaborate across teams to deliver measurable improvements on NVIDIA GPUs.

You will profile using Nsight tools, explore batching, quantization, and MoE serving, and drive downstream patches that advance the public Pareto frontier while ensuring strict correctness.

Qualifications

• Advanced degree or equivalent in CS/CE/EE or related field.
• 3+ years of relevant engineering experience.
• Proven knowledge of AI model execution optimization, batching, MoE, quantization.
• Experience benchmarking GPU workloads with Nsight tools.
• Strong Python and C++/CUDA skills.
• Ability to communicate performance tradeoffs clearly.

Responsibilities

• Distill performance instincts into reusable workflows and configurations for autonomous AI agents.
• Improve throughput-latency tradeoffs in AI inference workloads and explore batching, parallelism, and quantization options.
• Measure and optimize serving architectures across TensorRT-LLM, vLLM, and Dynamo on NVIDIA GPUs.
• Profile workloads with Nsight Systems/Compute and interpret kernel-level data to drive fixes.
• Land upstream improvements such as patches and deployment recipes while maintaining model correctness.

Skills

Python

C++/CUDA

Experiment design

Profiling

Technical communication

Education

BS/MS/PhD in CS/CE/EE or related

Tools

Nsight Systems

Nsight Compute

CUPTI

PyTorch profiler

Job description

NVIDIA is seeking a Senior Inference Performance Engineer to push performance limits for large-scale AI inference. You will optimize AI model execution, distill insights into reusable methods, and collaborate across teams to deliver measurable improvements on NVIDIA GPUs.

You will profile using Nsight tools, explore batching, quantization, and MoE serving, and drive downstream patches that advance the public Pareto frontier while ensuring strict correctness.