Principal High-Performance LLM Training Engineer

NVIDIA•Santa Clara, CA

8d•$272,000 - $431,250

About The Position

NVIDIA is seeking a Principal Engineer to drive the performance of large-scale AI training and post-training workloads across its hardware and software stack. The role spans distributed training, GPU architecture, systems software, deep learning frameworks, and performance engineering. The engineer will analyze and optimize frontier-scale LLM workloads running on thousands of GPUs, improve frameworks including PyTorch, JAX, NeMo, and NeMo RL, and use workload insights to inform future NVIDIA GPU, system, and software roadmaps. The role calls for a deeply technical leader who can work across application-level training behavior, framework and runtime internals, CUDA libraries, communication collectives, memory systems, networking, and GPU architecture. The Principal Engineer will directly improve performance, set technical direction, raise organizational standards, and influence cross-functional decisions.

Requirements

MS or PhD, or equivalent experience, in Computer Science, Electrical Engineering, Computer Engineering, or a related field.
12+ years of relevant work or research experience.
Demonstrated principal-level technical impact in one or more of these areas: large-scale AI training systems, GPU performance optimization, distributed systems, high-performance computing, ML frameworks, compilers/runtimes, or hardware/software co-design.
Deep hands-on experience analyzing and optimizing large-scale deep learning workloads, especially transformer-based models, LLM pre-training, reinforcement learning, fine-tuning, or other post-training workloads.
Strong understanding of GPU and AI accelerator architecture, from individual accelerators to datacenter-scale systems.
Experience with distributed training techniques such as data, tensor, pipeline, expert, and sequence parallelism; activation checkpointing; mixed-precision training; and communication/computation overlap.
A strong track record of using profiling, tracing, benchmarking, and performance-modeling tools to diagnose complex bottlenecks and drive measurable improvements.
Excellent communication and technical leadership skills, including the ability to influence architecture and software decisions across multiple teams without direct authority.

Responsibilities

Lead end-to-end performance analysis and optimization of innovative LLM pre-training and post-training workloads on NVIDIA hardware and software platforms.
Identify and remove bottlenecks across compute, memory, communication, scheduling, parallelism strategy, kernel efficiency, framework overhead, and system-level scaling.
Develop production-quality software, tools, models, benchmarks, and analysis infrastructure to improve training performance, efficiency, and developer velocity across NVIDIA’s AI software stack.
Build and refine performance models, workload characterizations, and simulation methodologies to guide GPU, networking, system, and software architecture decisions.
Serve as a technical authority for AI training performance and partner with teams across GPU architecture, systems, CUDA libraries, compilers, networking, frameworks, product management, and applied AI.
Translate workload insights into hardware and software recommendations and advocate for changes that improve performance and efficiency across the AI ecosystem.
Mentor engineers and provide technical leadership, helping establish best practices for large-scale AI performance analysis and optimization.