AI Inference Performance Engineer

NVIDIA•Santa Clara, CA

1d•Hybrid

About The Position

We optimize and benchmark GenAI inference on NVIDIA's latest accelerators, defining the industry’s performance standards across language models, video generation, and speech workloads. We work directly within TensorRT-LLM, SGLang, and vLLM, building the tools that evaluate serving performance at scale. This team sits at the intersection of GPU performance engineering and public accountability. What You Will Be Doing: Drive industry benchmark results: own the end-to-end optimization pipeline, implement and integrate optimizations in quantization, scheduling, memory management, and distributed inference across TensorRT-LLM, SGLang, and vLLM. Define and optimize cutting-edge workloads: identify and shape next-generation inference benchmarks, multi-turn coding, agentic workflows, and other emerging AI use cases. Collaborate with framework and kernel teams to push performance to its extreme on large-scale LLM-MoE models, vision-language models, video diffusion models, recommendation, and speech workloads. Architect distributed inference: Design and optimize execution from single-GPU to rack-scale clusters, managing performance across clusters of GPUs. Establish performance methodology: Apply roofline analysis and systematic profiling to decompose bottlenecks across CUDA kernels, frameworks, and serving layers. Influence the ecosystem: contribute to TensorRT-LLM, vLLM, SGLang, and other open-source projects. Partner with architecture, kernel, and compiler teams to shape GPU roadmaps based on real workload data. Technical Leadership: Raise the technical bar for the team, drive cross-functional execution on tight benchmark timelines, and lead a world-class team. GPU deep learning has provided the foundation for machines to learn, perceive, reason and solve problems posed using human language. The GPU started out as the engine for simulating human imagination, conjuring up the outstanding virtual worlds of video games and Hollywood films. Now, NVIDIA's GPU runs deep learning algorithms, simulating human intelligence, and acts as the brain of computers, robots and self-driving cars that can perceive and understand the world. Just as human imagination and intelligence are linked, computer graphics and artificial intelligence come together in our architecture. Two modes of the human brain, two modes of the GPU. This may explain why NVIDIA GPUs are used broadly for deep learning, and NVIDIA is increasingly known as “the AI computing company.” Come, join our DL Architecture team, where you can help build the real-time, cost-effective computing platform driving our success in this exciting and quickly growing field.

Requirements

BS, MS, or PhD in Computer Science, Computer Engineering, Electrical Engineering, or equivalent experience.
5+ years of relevant software development experience.
Strong Python or C++ programming, software design, and software engineering skills.
Expertise with a DL framework such as PyTorch or JAX.
Proven track record of delivering measurable performance improvements in deep learning inference or high-performance systems.
Deep understanding of LLM/VLM architectures and inference mechanics: attention, KV caching, batching strategies, decode-phase bottlenecks, speculative decoding, disaggregated serving etc.

Nice To Haves

Prior experience with an LLM framework (TensorRT-LLM, vLLM, SGLang, etc) or a DL compiler in inference, deployment, algorithms, or implementation.
Prior experience with performance modeling, profiling, debug, and code optimization of a DL/HPC/high-performance application.
Experience with scale-out inference orchestration (MPI, NCCL, K8S) on large GPU clusters.
Expertise in kernel development (CUTLASS, cuteDSL, tilelang, OpenAI Triton) or compiler/runtime paths (torch.compile, graph lowering, operator fusion).
Architectural knowledge of CPU, GPU, FPGA or other DL accelerators; GPU programming experience (CUDA).
Track record of leading ambiguous, high-impact technical programs across multiple teams under tight deadlines.

Responsibilities

Drive industry benchmark results: own the end-to-end optimization pipeline, implement and integrate optimizations in quantization, scheduling, memory management, and distributed inference across TensorRT-LLM, SGLang, and vLLM.
Define and optimize cutting-edge workloads: identify and shape next-generation inference benchmarks, multi-turn coding, agentic workflows, and other emerging AI use cases.
Collaborate with framework and kernel teams to push performance to its extreme on large-scale LLM-MoE models, vision-language models, video diffusion models, recommendation, and speech workloads.
Architect distributed inference: Design and optimize execution from single-GPU to rack-scale clusters, managing performance across clusters of GPUs.
Establish performance methodology: Apply roofline analysis and systematic profiling to decompose bottlenecks across CUDA kernels, frameworks, and serving layers.
Influence the ecosystem: contribute to TensorRT-LLM, vLLM, SGLang, and other open-source projects.
Partner with architecture, kernel, and compiler teams to shape GPU roadmaps based on real workload data.
Raise the technical bar for the team, drive cross-functional execution on tight benchmark timelines, and lead a world-class team.