AMD GPU Monitor

AMD GPU Monitor: Monitor AMD Radeon GPU Usage and VRAM on Linux

Monitoring an AMD Radeon graphics card on Linux can be challenging when you need more than basic GPU utilisation statistics. Whether you run artificial intelligence models, machine learning workloads, scientific simulations, or graphics applications, understanding how your GPU uses video memory (VRAM) is essential for diagnosing performance problems and making the most of your hardware.

AMD GPU Monitor is a Python-based Linux monitoring tool designed to provide a detailed view of AMD GPU activity in a single terminal interface. It combines VRAM monitoring, per-process GPU memory usage, GPU utilisation, power consumption, temperature monitoring, historical statistics, and diagnostic warnings.

Instead of presenting disconnected hardware measurements, AMD GPU Monitor helps explain which applications are using your graphics card, how much video memory they consume, and whether your system is approaching its hardware limits.

Monitor AMD Radeon GPU VRAM Usage by Process

One of the most useful features of AMD GPU Monitor is its detailed breakdown of video memory usage. Knowing that an AMD Radeon graphics card has almost exhausted its available VRAM is useful, but identifying which applications are consuming that memory is even more valuable.

Consider an AMD Radeon RX 7900 XTX with 24 GiB of dedicated video memory. When running a large language model through Ollama, AMD GPU Monitor can distinguish between memory used by the main computational workload and memory associated with the desktop environment and graphics driver.

For example, a monitoring session might show the following:

• 22.2 GiB used by the main ROCm compute workload.
• 667 MiB attributed to graphics and driver memory.
• 1.2 GiB of available VRAM.

This breakdown helps explain why the GPU is nearly full and how much memory remains available for additional applications. A large language model may deliberately occupy most of the available VRAM, while the desktop environment still requires memory for its own operations.

The process table identifies applications by their process names and PIDs and displays estimated VRAM residency, evicted memory, and relative memory usage. This makes it easier to investigate unexpected GPU memory consumption without relying exclusively on total memory counters.

The tool also distinguishes between memory resident in VRAM and memory that prefers VRAM but has been evicted elsewhere. This can be particularly useful when investigating graphics applications and desktop compositors.

Memory accounting depends on the counters exposed by the Linux kernel and AMDGPU driver. Some per-process values are estimates, and shared graphics resources may be counted for multiple processes, so the sum of individual process figures may not match the card's total VRAM usage.

SAMD Radeon GPU Monitoring: Utilisation, Temperature and Power

AMD GPU Monitor provides more than an AMD Radeon VRAM usage monitor. It brings together several important hardware measurements to help users understand how their graphics card behaves under load.

The interface can display:

• GPU utilisation: Current GPU activity and historical averages.
• Power consumption: Measured power alongside the reported power limit, where available.
• GPU edge temperature: The main GPU temperature.
• Hotspot temperature: The hottest measured point on the GPU.
• Memory temperature: The reported temperature of the video memory.
• VRAM usage: Dedicated video memory consumption.
• GTT usage: System memory mapped for GPU use, including relevant GPU buffers.

Historical statistics allow users to examine current GPU activity alongside short-term and longer-term averages. This is useful when investigating fluctuating GPU loads, sustained computational workloads, rising temperatures, or unexpectedly high power consumption.

For example, a GPU utilisation reading of 100% indicates that the GPU is busy, but it does not explain which application is responsible. Combining utilisation statistics with per-process memory accounting provides additional context for understanding what is happening on the system.

Sensor availability and interpretation depend on the GPU model, driver, kernel, and system configuration.

Identify GPU Memory Problems Before They Affect Your Workload

When VRAM becomes nearly full, starting another GPU-intensive application may fail, or an existing workload may become constrained by memory availability. AMD GPU Monitor helps identify these situations by displaying available VRAM and issuing contextual warnings when memory headroom becomes low.

The status section can also identify the largest graphics-memory consumer and report evicted graphics memory. Rather than requiring users to interpret several independent statistics, the tool presents related information together to help investigate potential problems.

This is particularly useful for Linux workstations running:

• Large language models and local AI applications.
• Ollama and other inference software.
• ROCm and GPU-accelerated machine learning workloads.
• Scientific simulations and mathematical calculations.
• Graphics applications and GPU-intensive desktop workloads.

By showing which processes consume GPU memory and how the available capacity is changing, AMD GPU Monitor can help users investigate resource shortages and make better-informed decisions about their workloads.

AMD GPU Monitor vs NVTOP, ROCm SMI and RadeonTOP

Several existing utilities provide GPU monitoring on Linux, including NVTOP, ROCm SMI, and RadeonTOP. Each offers useful functionality, but users may need a different combination of measurements depending on whether they are investigating GPU performance, video memory consumption, hardware sensors, or individual processes.

NVTOP

NVTOP provides an interactive interface for monitoring supported GPUs and their processes. Its AMD-specific capabilities depend on the software version, kernel, driver, and available monitoring interfaces.

Some users may encounter driver-specific bugs. For example, an assertion failure in the AMDGPU process-information code can cause NVTOP to terminate in a particular environment. This illustrates a potential issue with a particular software configuration rather than establishing that NVTOP is generally unstable.

ROCm SMI

ROCm SMI provides hardware management and monitoring functionality for supported AMD GPUs. It is useful for accessing hardware information, but users who want an integrated terminal display combining process-level memory accounting, historical readings, and contextual warnings may prefer a different presentation.

RadeonTOP

RadeonTOP provides detailed information about AMD GPU activity and graphics-engine utilisation. Its emphasis on hardware statistics can be valuable for low-level investigation, although additional context may be needed when diagnosing specific VRAM or application-level memory problems.

What makes AMD GPU Monitor different?

AMD GPU Monitor focuses on bringing several aspects of GPU monitoring together in one interface:

• Detailed VRAM accounting: Separates compute-related memory from graphics and driver memory where the available counters permit it.
• Per-process monitoring: Shows which applications are using GPU memory.
• Evicted memory reporting: Helps identify graphics memory that prefers VRAM but is currently placed elsewhere.
• Historical measurements: Presents current readings alongside averages over different time intervals.
• Hardware sensors: Combines utilisation, power, and temperature information in one display.
• Diagnostic warnings: Highlights low VRAM availability and other potentially important conditions.

The key advantage is not necessarily that every individual measurement is unique. It is that the information is organised to help users understand what their GPU is doing and investigate why.

The capabilities of all these utilities depend on the GPU model, Linux kernel, driver, and software environment. AMD GPU Monitor is designed to provide a more integrated and informative monitoring experience rather than replace every specialised diagnostic tool.

AMD GPU Monitoring on Linux and Ubuntu

AMD GPU Monitor is written in Python and is intended for Linux systems using AMD GPUs. It obtains information from the Linux kernel, AMDGPU driver, and available hardware monitoring interfaces.

The project aims to support a broad range of AMD Radeon graphics cards and Linux environments. The availability of individual measurements depends on the interfaces exposed by the installed kernel and driver, so some features may vary between systems.

For Ubuntu users, AMD GPU Monitor provides a practical way to investigate AMD Radeon GPU utilisation, video memory consumption, process-level memory usage, and hardware sensor readings from the terminal. The same approach is useful on other Linux distributions that provide the required monitoring interfaces and Python environment.

How to Run AMD GPU Monitor

The source code and technical documentation are available on GitHub:

AMD GPU Monitor — GitHub repository

To start the monitor with a refresh interval of one second, run the following command from the directory containing the script:

sudo python amd_gpu_monitor.py -r 1

The -r 1 option specifies a one-second refresh interval. Elevated privileges may be required to access all process and GPU information, depending on the system configuration.

Consult the project's README for installation instructions, available command-line options, technical details, and information about monitoring features.

Conclusion: A Detailed AMD Radeon GPU and VRAM Monitor

AMD GPU Monitor is designed for Linux users who need more than a simple GPU utilisation percentage. By combining VRAM monitoring, per-process memory usage, temperature and power measurements, historical statistics, and diagnostic warnings, it provides a detailed view of AMD Radeon GPU activity in one terminal interface.

Whether you are monitoring an AMD Radeon RX 7900 XTX, investigating GPU memory consumption in a local AI setup, running ROCm workloads, or troubleshooting unexpected VRAM usage on Ubuntu, AMD GPU Monitor helps connect hardware statistics with the processes responsible for them.

Visit AMD GPU Monitor on GitHub to explore the source code, read the documentation, and use the tool to monitor AMD Radeon GPU performance and video memory usage on Linux.

Published: 2026-10-18 18:38:51