Senior Systems Performance Engineer - HPC & GPU

NVIDIA Corporation; United States; Remote; USD 150,000 - 220,000; Full time

Job summary

NVIDIA Corporation in the United States seeks a proficient engineer for its Applied Systems Engineering Performance team. This role blends experimentation and software engineering to quantify, diagnose, and improve HPC and AI workloads on GPU-accelerated platforms.

You will develop repeatable experiments, analyze performance across nodes and interconnects, and collaborate with hardware, firmware, and software teams to deliver actionable improvements that scale at datacenter level.

Qualifications

• Extensive practical experience in Linux performance analysis with focus on processes, threads, NUMA and I/O.

Responsibilities

• Characterize node, rack, and cluster performance with microbenchmarks and representative apps.
• Identify Linux performance problems related to CPU scheduling, NUMA locality, affinity, interrupts, and contention.
• Examine distributed workload scaling and impact of topology on performance.
• Combine traces, counters and telemetry to isolate root causes and propose fixes.
• Develop Python tools for automated benchmark execution and data analysis.
• Define and validate performance improvements for new platforms and deployments.

Skills

Linux performance analysis

Python scripting

C/C++/CUDA performance

HPC systems

Profiling and tracing

Education

BS or MS in Computer Science/Engineering/Math/Physics

Tools

Nsight Systems

Nsight Compute

Linux perf

eBPF tooling

MPI/NCCL/RDMA

Job description

NVIDIA Corporation in the United States seeks a proficient engineer for its Applied Systems Engineering Performance team. This role blends experimentation and software engineering to quantify, diagnose, and improve HPC and AI workloads on GPU-accelerated platforms.

You will develop repeatable experiments, analyze performance across nodes and interconnects, and collaborate with hardware, firmware, and software teams to deliver actionable improvements that scale at datacenter level.