Senior AI Inference Performance Engineer

NVIDIA AI; Santa Clara (CA); On-site; USD 140,000 - 230,000; Full time

Job summary

NVIDIA is seeking a Senior Inference Performance Engineer to push performance limits on large-scale AI inference benchmarks using modern GPUs. You will optimize autonomous benchmarking and profiling workflows, enabling agents to run experiments, tune configurations, and report credible wins.

You will collaborate across TensorRT-LLM, vLLM, and CUDA teams, exercise rigorous experiments, and shape patches to serving frameworks.

Qualifications

• Experience optimizing AI model execution for throughput-latency tradeoffs.
• Hands-on benchmarking with Nsight Systems, Nsight Compute, CUPTI, or PyTorch profiler.
• Strong Python skills and ability to modify large C++/CUDA codebases.
• Rigorous experimental methodology with reproducible benchmarks.
• Experience with quantization and KV cache considerations.
• Ability to explain tradeoffs clearly in writing and speech.

Responsibilities

• Distill performance instincts into reusable skills and methodologies for autonomous AI agents.
• Improve AI inference workloads to increase throughput-per-GPU and user interactivity.
• Measure and optimize serving architectures across TensorRT-LLM, SGLang, vLLM, and Dynamo.
• Profile workloads with Nsight tools and analyze results to drive fixes.
• Land upstream improvements to serving frameworks and kernels.
• Collaborate with TensorRT-LLM, SGLang, vLLM, kernel, benchmarking, and GPU teams.

Skills

Python

C++/CUDA

Nsight profiling

PyTorch profiler

GPU benchmarking

MoE serving

Quantization

Kernel optimization

Kernel profiling

Education

BS/MS/PhD in CS/CE/EE

Tools

Nsight Systems

Nsight Compute

CUPTI

TensorRT-LLM

vLLM

Job description

NVIDIA is seeking a Senior Inference Performance Engineer to push performance limits on large-scale AI inference benchmarks using modern GPUs. You will optimize autonomous benchmarking and profiling workflows, enabling agents to run experiments, tune configurations, and report credible wins.

You will collaborate across TensorRT-LLM, vLLM, and CUDA teams, exercise rigorous experiments, and shape patches to serving frameworks.