When a game feels slow, the first instinct is often to look for a single FPS number. That number is useful, but it does not explain why the game feels bad. Linux gamers can use MangoHud, an open-source overlay for Vulkan and OpenGL applications, to observe frame rate, frame times, CPU and GPU load, temperatures, and other metrics.

The goal is not to cover the screen with numbers. It is to use a few measurements to answer a specific question: Is the game limited by the GPU? Is frame delivery uneven? Is the CPU saturated? Is the system overheating or simply waiting on a shader compilation? MangoHud can help collect clues, but it cannot diagnose every cause automatically.

What MangoHud displays

MangoHud is an overlay that can show performance information while a game runs. Its project documentation lists data such as FPS, frame-time information, CPU and GPU usage, and temperatures. The exact metrics available depend on hardware, drivers, permissions, and the installed build.

On Fedora, for example, the project’s current README documents installation through the distribution package manager with sudo dnf install mangohud. Other distributions use their own package names and repositories. If you use Steam as a Flatpak, there is a Flatpak runtime extension for MangoHud, and the setup differs from a system package. Follow the instructions for the Steam packaging method you actually use.

For a normal Steam game on a system-installed build, the basic launch option is commonly:

mangohud %command% 

For some setups, an environment variable can also enable the layer, but the project notes differences between Vulkan, OpenGL, Gamescope, and Flatpak configurations. Start with the documentation for your build rather than assuming one command works everywhere.

Read frame time, not only FPS

FPS is the number of frames presented per second over a period. Frame time measures how long each frame takes. At 60 FPS, the average frame interval is about 16.7 milliseconds. At 30 FPS, it is about 33.3 milliseconds. These averages do not show whether individual frames arrive evenly.

A game can report a high average FPS while producing occasional long frames that look like stutter. A frame-time graph can reveal spikes that the average hides. If the graph is mostly consistent and then spikes during a new area or effect, shader compilation or asset streaming may be involved. If the frame times remain high throughout a scene, the game may be under sustained CPU or GPU load.

When comparing settings, use the same reproducible segment. A busy city, menu, and empty corridor can put different loads on the system. Record the same route or benchmark, use the same resolution and graphics settings, and compare more than a short moment.

Use CPU and GPU load as clues

If the GPU stays near full utilization while the frame rate is below your target, the game may be GPU-bound. Reducing resolution or expensive graphics settings can test that theory. If GPU use is low while one or more CPU cores are heavily loaded, the game may be CPU-bound or waiting on another part of the system.

These indicators are not absolute. Some games distribute work unevenly across CPU cores. Driver overhead, synchronization, storage stalls, and frame caps can all make utilization look lower than expected. A game capped at 60 FPS may not use the GPU fully because it does not need to. Low usage is not automatically evidence of a broken driver.

Temperatures and clock speeds add context. If performance drops after several minutes while temperatures rise, thermal or power limits may be involved. But a high temperature by itself is not a diagnosis. Compare the behavior to the device’s expected operating range and check system logs or vendor guidance before changing fan curves or power limits.

A repeatable measurement workflow

Start with one question. For example: “Does lowering the resolution improve performance?” Launch the game with the same scene and collect a baseline. Change the resolution only, repeat the scene, and compare frame-time consistency, average FPS, GPU load, and temperature.

Then test another variable if needed. If lowering resolution changes little, the GPU may not be the limiting factor. Try a graphics setting that affects CPU or GPU work, or check whether the game’s frame limiter is active. Change one thing at a time and write down the result.

A useful note might read: “In the same 90-second route, 1440p medium averages 54 FPS with repeated 30–40 ms spikes; 1080p medium averages 56 FPS with similar spikes; GPU use falls from 97% to 72%.” This suggests that resolution was not the only factor, though more investigation is needed. It is more actionable than “performance feels bad.”

Common measurement traps

Do not compare one run with shader caches warmed against another first launch without noting the difference. Shader compilation can create temporary stutter, and the behavior may change on a later run. Similarly, background updates, browser tabs, recording software, and overlays can affect results.

A frame cap changes utilization. If a game is capped, a GPU may sit below full load by design. Disable or account for V-sync, in-game limiters, driver-level caps, Gamescope limits, and refresh rate when interpreting results. Multiple overlays or capture tools can also interfere with one another.

Overlay data may be incomplete. Some sensors require additional permissions or libraries. A missing temperature reading does not mean the GPU has no temperature sensor; it may mean the overlay cannot access it. Do not run the entire game as root to make an overlay show a metric.

Gamescope and MangoHud

Gamescope can run a game in a nested compositor, and MangoHud can display telemetry. The projects have a specific integration path: MangoHud’s documentation indicates using mangoapp with Gamescope when appropriate, rather than assuming the ordinary MangoHud injection works in every Gamescope configuration. Check current project documentation and distro packaging because options evolve.

Avoid piling Gamescope, MangoHud, multiple frame limiters, and launch variables into a single command before testing each component. First confirm the game works. Then enable the overlay by itself. Add Gamescope separately if you need its scaling or window behavior. This isolates errors and makes results easier to interpret.

When to share a performance report

Include hardware, distribution, kernel and graphics driver versions, game version, Proton version if applicable, resolution, settings, frame cap, and the scene tested. Note whether the figures are from a cold or warmed-up run. If you post screenshots, remove usernames or other private information visible in the overlay or desktop.

MangoHud is not a benchmark suite and its metrics do not tell the whole story. Input latency, display refresh, frame pacing, game engine behavior, and subjective smoothness still matter. Use the overlay to support a controlled comparison, not to chase a single number.

A good performance investigation starts with a question, collects measurements under repeatable conditions, and changes one setting at a time. MangoHud makes more of the system visible while you play. The real skill is interpreting those signals carefully enough to know which layer to test next.

References