Exam NCP-AAI Topic 1 Question 91 Discussion

End-to-end latency waterfalls show where time is spent across guardrails, queues, and inference. Local component tuning misses cross-service overhead. The correct implementation surface is profiling the request path from ingress through guardrails, routing, Triton scheduling, TensorRT-LLM execution, and response assembly. The selected option specifically C states "Create end-to-end latency waterfalls that capture guardrail overhead, NIM queuing delays, and TensorRT optimization benefits while assessing overall pipeline efficiency.", which matches the operational requirement rather than a superficial wording match. From an NVIDIA systems-engineering lens, Option C aligns with the way agentic services should be decomposed and measured. The alternatives would look simpler in a prototype, but overlarge batches may improve throughput while violating interactive latency targets. The NVIDIA implementation angle is not cosmetic here: NVIDIA Perf Analyzer, GenAI-Perf, Nsight, and Triton metrics help isolate whether the bottleneck is batching, compute, memory, or request scheduling. This choice gives engineering teams the knobs they need for continuous tuning after deployment.