To view the usage of NVIDIA GPUs on Linux, the preferred tool is nvidia-smi. 1. Run nvidia-smi directly to view GPU usage, memory usage, temperature and other information; 2. Use watch -n 1 nvidia-smi to achieve real-time monitoring; 3. Obtain structured data through --query and --format parameters, such as nvidia-smi --query-gpu=index,name,temperature.gpu,utilization.gpu --format=csv; 4. Non-NVIDIA graphics cards can use rocm-smi (AMD) or intel_gpu_top (Intel); 5. Graphical interface tools such as nvtop and gpustat provide a more intuitive monitoring experience.

Want to know how to see GPU usage on Linux? If you are using an NVIDIA graphics card, it is actually quite convenient. The system comes with its own or third-party tools to help you quickly check the current GPU usage, temperature, video memory usage, etc. The following methods are more practical and are suitable for daily monitoring or debugging programs.

View NVIDIA GPU usage: nvidia-smi

This is the most direct method and the preferred command line tool for most Linux users. After installing the NVIDIA driver, the system will come with this command.

The operation mode is very simple, enter:

 nvidia-smi

You will see a table showing the usage rate (GPU%), memory usage (Used / Total), temperature, power status and other information per GPU. The default is static output once. If you want to monitor in real time, you can add -q or cooperate with the watch command:

 watch -n 1 nvidia-smi

This way, refreshes once a second and you can see dynamic changes. It is suitable for running training tasks to see if there are any free cards or whether a process is full of resources.

Get more detailed data: use nvidia-smi query mode

If you need data in a specific field, such as just looking at the memory usage of a certain card, you can use --query with --format to get structured output. For example:

 nvidia-smi --query-gpu=index,name,temperature.gpu,utilization.gpu --format=csv

This outputs GPU number, name, temperature and usage in CSV format, making it easier to script calls or log records.

Common queries include:

• index : GPU number
• temperature.gpu : temperature
• utilization.gpu : GPU usage rate
• memory.used , memory.total : memory usage and total amount

What to do if you don’t have a graphics card?

If you are not using an NVIDIA graphics card, such as AMD or Intel, you need to change to another tool. AMD can use rocm-smi , and Intel graphics cards can use intel_gpu_top (intel-gpu-tools is required to be installed).

The functions of these tools are similar to nvidia-smi , and you can also view usage, frequency, temperature and other information, but the command format is slightly different. It is recommended to go to the corresponding official website to find the documents to see the basic usage.

Graphical interface monitoring tool

Don't want to type the order? You can also use the graphical interface tool to view it. Tools like nvtop or gpustat provide a more intuitive display.

For example, nvtop is modeled on the top command style, supports real-time viewing of all GPU status, and can also see the usage of GPUs of each process. It is not complicated to install. It can generally be compiled and installed through the package manager or from the source code.

gpustat is a Python tool that can briefly display the GPU status in the terminal, suitable for a quick glance on a remote server.

Basically these are the methods. Different tools can be selected for different scenarios. nvidia-smi is the most basic and commonly used, and other methods can be used according to needs.

The above is the detailed content of how to check gpu usage linux. For more information, please follow other related articles on the PHP Chinese website!