An operations engineer wants to feed real-time GPU metrics—such as thermal conditions, power usage, and memory utilization—into a Prometheus and Grafana observability stack. Which NVIDIA utility should be deployed on the cluster nodes to export this telemetry?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Now, if you look at enterprise monitoring, Prometheus and Grafana are the industry favorites. But Prometheus can't just talk to a raw NVIDIA GPU out of the box. We need a translator. That translator is the NVIDIA DCGM Exporter. DCGM stands for Data Center GPU Manager. It's a suite of tools that sits right on the node and talks directly to the GPU drivers to collect all kinds of telemetry—things like temperature, fan speeds, power draw, and memory usage. The Exporter takes all that raw telemetry, reformats it into a clean Prometheus text format, and spins up a web endpoint that Prometheus can scrape. Pay close attention here: this is an exam favorite! Make sure you don't confuse this with Nsight Systems (which is a profiling tool you use to optimize code), NCCL (which is the library that helps GPUs talk to each other), or cuDNN (which contains the math primitives for deep learning). The telemetry export job belongs to DCGM Exporter.
Full explanation below image
Full Explanation
NVIDIA Data Center GPU Manager (DCGM) Exporter is a specialized tool designed to gather GPU telemetry and expose it to monitoring systems. DCGM itself is a suite of tools for managing and monitoring NVIDIA GPUs in clustered environments, tracking metrics such as GPU utilization, memory usage, temperature, power consumption, clock speeds, and hardware errors (like ECC memory errors). The DCGM Exporter hooks into the DCGM engine, translates these metrics into a format compliant with Prometheus standards, and exposes them via an HTTP endpoint. This allows Prometheus servers to scrape the metrics at regular intervals and forward them to visualization tools like Grafana. Let's analyze the incorrect options: - Option A is incorrect because NVIDIA Nsight Systems is a system-wide profiling tool used by developers to analyze application performance, identify GPU bottlenecks, and optimize execution paths, rather than a telemetry exporter for continuous monitoring. - Option B is incorrect because NVIDIA Collective Communications Library (NCCL) provides multi-GPU and multi-node collective communication primitives (such as All-Reduce, All-Gather, Broadcast) optimized for NVIDIA GPUs, and is not involved in system monitoring. - Option D is incorrect because NVIDIA CUDA Deep Neural Network library (cuDNN) is a GPU-accelerated library of primitives for deep neural networks, providing implementations for routines like convolution, pooling, normalization, and activation functions, rather than a system-level telemetry exporter.