Senior LLM Training Performance Architect

NVIDIA Corporation; Santa Clara (CA); On-site; USD 272,000 - 431,000; Full time

Benefits offered by this job

Equity

Benefits

Job summary

NVIDIA Corporation seeks a Principal Engineer to drive the performance of large-scale AI training and post-training workloads across its hardware and software stack. You will work across GPU architecture, systems software, and deep learning frameworks to remove bottlenecks and guide roadmaps.

You will mentor engineers, build production tools, and shape performance benchmarks for next-generation AI workloads, collaborating with multiple teams to raise the organization’s performance bar.

Qualifications

• MS or PhD (or equivalent) in Computer Science, Electrical Engineering, Computer Engineering, or related field.
• 12+ years of relevant work or research experience.
• Deep hands-on experience optimizing performance of large-scale deep learning workloads.

Responsibilities

• Lead end-to-end performance analysis and optimization of LLM pre-training and post-training workloads on NVIDIA hardware and software.
• Identify bottlenecks across compute, memory, communication, and scheduling to improve training speed.
• Develop production-quality software, tools, models, and benchmarks for performance and efficiency.
• Mentor engineers and influence architecture and software decisions across teams.

Skills

Large-scale AI training

GPU performance optimization

Distributed systems

Profiling & benchmarking

Technical leadership

Education

MS/PhD in CS/EE or related field

Tools

PyTorch

JAX

NeMo/NeMo RL

Job description

NVIDIA Corporation seeks a Principal Engineer to drive the performance of large-scale AI training and post-training workloads across its hardware and software stack. You will work across GPU architecture, systems software, and deep learning frameworks to remove bottlenecks and guide roadmaps.

You will mentor engineers, build production tools, and shape performance benchmarks for next-generation AI workloads, collaborating with multiple teams to raise the organization’s performance bar.