Your question is Nsight for Inference Profiling. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
Tell me about your experience using NVIDIA Nsight tools to speed up inference.
Explain a hands-on workflow using NVIDIA Nsight Systems and NVIDIA Nsight Compute. Cover how you located bottlenecks across Triton Inference Server, TensorRT or TensorRT-LLM, CUDA kernels, Dynamic Batching, KV Cache behavior, and GPU or CPU interaction. Discuss the baseline, profiling evidence, changes made, validation method, and any accuracy, latency, throughput, or cost tradeoffs.