GPU Performance & Profiling Engineer

Ensigncode is hiring a GPU Performance & Profiling Engineer to own measurement across our GPU optimization team: building reliable benchmark suites and using Nsight Systems and Nsight Compute to diagnose compute, memory-bandwidth, launch, and synchronization bottlenecks. You'll independently validate claimed improvements and turn performance data into priorities for the rest of the team.

Every optimization effort on our GPU team starts with profiling, not guesswork. As GPU Performance & Profiling Engineer, you’ll own that measurement discipline: building reproducible benchmark suites, running Nsight Systems for system-wide timeline analysis and Nsight Compute for per-kernel metrics, and turning raw data into a clear, prioritized list of what’s actually worth fixing.

You’ll need a measurement-first mindset and the diagnostic instinct to tell the difference between measurement noise and a real bottleneck. Your reports are what the kernel and inference engineers act on, and you’re also the one who independently confirms that a claimed speedup actually holds up.

Responsibilities

• Build reliable benchmark suites and reproducible performance experiments across workloads and GPU configurations
• Profile using Nsight Systems and Nsight Compute; identify compute, memory-bandwidth, launch, synchronization, and pipeline bottlenecks
• Analyze occupancy, warp stalls, memory transactions, utilization, and end-to-end latency/throughput
• Create performance reports, regression checks, and actionable optimization recommendations

Core requirements

• CUDA fundamentals; profiling tools; benchmarking methodology; performance analysis and scripting (Python or C++)
• Strong diagnostic thinking; ability to separate measurement noise from real gains

Team contribution

Turns performance data into priorities for kernel and inference engineers; independently validates claimed improvements.

Works most closely with