Monitoring Nvidia GPUs using REST API

In this article I will tell how can get information about the state of the Nvidia GPU through the REST API.

For work, let’s take https://github.com/lampaa/nvidia-smi-rest. SmiRest works through nvidia-smi, receives all information about the GPU and outputs it in json format (more about the format here):

JSON ответ на запрос /v1

After launch, the REST API is available at http://localhost:8176

Get information about the version of the driver and cuda, a list of all GPUs http://localhost:8176/v1

Get information about one GPU (/v1/GPU_UUID): http://localhost:8176/v1/GPU-a9685d8a-fbf2-7465-ee1f-307141ef06a8

Get information about a specific field of one GPU: (/v1/GPU_UUID/field[/subfield][/subsubfield]): http://localhost:8176/v1/GPU-a9685d8a-fbf2-7465-ee1f-307141ef06a8/pci/pciGpuLinkInfo/pcieGen

The service has built-in statistics per minute and displaying statistics on charts.

Get statistics for all GPUs (/stats): http://localhost:8176/stats

Get statistics for one GPU (/stats/GPU_UUID): http://localhost:8176/stats/GPU-a9685d8a-fbf2-7465-ee1f-307141ef06a8

Display graphics for all GPUs (/stats/graphs): http://localhost:8176/stats/graphs

You can add a library for working with a nvidia-smi to your maven project:

<dependency>
    <groupId>com.github.lampaa</groupId>
    <artifactId>smirest</artifactId>
    <version>0.0.2</version>
</dependency>

and get data:

package com.github.lampaa.smi;

import com.github.lampaa.smi.dtoV1.NvidiaSmiLogType;
import jakarta.xml.bind.JAXBException;
import java.io.IOException;

class Application {
    @Test
    public void void readingTest() throws JAXBException {
        //read from system
        NvidiaSmiLogType smiLog = SmiReader.read();
    }
}