Four games on CachyOS and Strix Halo

Four games, one run shown per game

These are current-settings personal snapshots, not repeated controlled tests. Rocket League has two valid runs; this display uses the one with the higher unrounded average FPS and never blends their measurements. All valid captures remain in the downloadable data.

Read the ARC Raiders result with its loading period. The full capture includes loading and two large frame gaps near visible loading/tip screens. Its 3.2 FPS calculated 1% low is not a sustained active-gameplay rate. The captured footage mainly shows traversal, not a demanding firefight.

RAM, VRAM and GTT are mean readings over approximately aligned capture windows. These accounting views overlap on a unified-memory machine; do not add them together.

FPS, pacing and memory

Open per-game frame-time timelines and system load

What the settings screens showed

ARC Raiders settings were observed after its capture. Overwatch settings were observed before its capture. These are documented screen states, not proof that every setting stayed fixed throughout play. Slay the Spire 2 and Rocket League settings were not completely recorded.

Definitions and limits

• Average FPS is 1,000 divided by the arithmetic mean of all recorded frame times in milliseconds.
• Calculated 1% low is 1,000 divided by the mean of the slowest 1% of frames, rounded up to a whole-frame count. This is not MangoHud's saved percentile-style “1% Min FPS.”
• P99 frame time is the 99th percentile threshold of recorded frame times. Frame time is not input latency or network ping.
• System RAM used is MemTotal minus MemAvailable from sysstat, including background activity. VRAM and GTT come from amdgpu_top. On this machine they are overlapping views, not separate physical capacity to add to RAM.
• Sensor alignment uses approximately one-second samples and the saved capture clock. It does not establish frame-exact causes for a spike.
• Scope: no matched Windows or Fedora baseline, no controlled repeat protocol, no proven game-only memory use, and no proof of maximum achievable FPS. Background load varied. Capture durations and scene mixes differ.

Sanitized measurements

The downloadable files preserve all five valid capture windows, including the second Rocket League run. These are derivatives of Jose's saved collector logs; original private run files and footage are not published.

• Metrics CSVOne row per valid capture; FPS, percentiles and sensor summaries
• Frame times CSVEvery valid frame sample in capture order
• Sensor series CSVApproximately aligned RAM, GPU and CPU intervals
• Public summary JSONDefinitions, run context, limitations and reported tool summaries
• Observed settings JSONTranscribed ARC Raiders and Overwatch menu values
• Presentation JSONThe four selected runs used in the slides

Sources and measurement tools

The FPS, frame-time and sensor numbers above are Jose's own September 2026 captures, calculated from the published frame-time samples and published derived sensor intervals. The linked tool documentation explains collection and memory terminology; it is not an independent validation of these personal results.

• MangoHud project and FPS logging documentation
• amdgpu_top GPU metrics project
• sysstat project for sar/sadf
• Linux kernel /proc memory definitions

The video and companion post discuss the experience of playing these games. This report keeps the measurements, definitions and limitations together for closer inspection.