Handheld Benchmark Metrics Explained: Average FPS, 1% Lows and Frametime
Learn what average FPS, 1% lows and frametime really reveal about handheld gaming performance—and how to compare benchmark results without being misled.
Published August 24, 2026Updated August 25, 2026
Category
Performance
Difficulty
Intermediate
Reading time
9 min
Published
August 24, 2026
HandheldAtlas guide
Complete walkthrough
Introduction
Two handhelds can report the same average frame rate and still feel completely different. One may deliver frames at a steady rhythm; the other may pause, hitch and surge. The difference is hidden in the timing.
This guide explains the three benchmark signals that matter most—average FPS, 1% lows and frametime—and shows how to read them together. It also sets out the test details that must match before two handheld results can be compared fairly.
Average FPS: useful, but incomplete
Average frames per second is a summary of rendering throughput across a test. It is useful for answering broad questions: did one preset produce more frames than another, and is performance near a target such as 30, 40 or 60 FPS?
The limitation is that an average compresses the whole run into one number. A sequence that alternates between fast frames and severe stalls can land on the same average as a smooth sequence. Average FPS therefore cannot prove that motion was consistent.
Use average FPS as the headline, not the whole story.
Frametime: the clearest view of smoothness
Frametime is the time required to present each frame, measured in milliseconds. Intel's performance guidance describes frame time as a more reliable way to discuss performance.
The conversion is:
• Frame time in milliseconds = 1000 ÷ FPS
• FPS = 1000 ÷ frame time in milliseconds
Common targets:
• 30 FPS = 33.33 ms
• 40 FPS = 25.00 ms
• 45 FPS = 22.22 ms
• 60 FPS = 16.67 ms
• 90 FPS = 11.11 ms
• 120 FPS = 8.33 ms
The budget is the time available for each frame. At 40 FPS, a stable result should stay close to 25 ms per frame. A frametime graph with a narrow, even trace usually feels consistent. Sharp upward spikes represent late frames and are perceived as hitches or stutter.
This is why a locked 40 FPS with stable 25 ms delivery can feel better than an unstable result that averages 50 FPS.
Official tools such as NVIDIA FrameView and AMD GPUOpen OCAT capture frame-rate and frame-time data. Both are designed to expose more than a single average.
What 1% Low FPS means
A 1% Low value summarizes the slowest part of a run without letting one isolated worst frame dominate the result. In the NVIDIA FrameView 1.9 user guide, 1% Low FPS is calculated by taking the slowest 1% of frames and averaging them.
If a benchmark averages 60 FPS but has a 1% Low of 32 FPS, the result contains meaningful slowdowns. If it averages 60 FPS with a 1% Low of 55 FPS, frame delivery is much more consistent. The gap between average FPS and 1% Low is therefore an important clue.
However, software does not always use identical percentile labels or calculations. Some tools report a low-FPS average, while others present percentile frametime statistics. Compare 1% Low figures only when the capture tool and method are known and consistent.
A missing 1% Low is not permission to estimate one. If the source provides only average FPS and a credible frametime graph, the 1% Low field should remain blank.
Why minimum FPS is not enough
Minimum FPS is the single lowest observed instant. It may be caused by a shader compilation event, loading transition, background task or capture anomaly. That makes it useful for spotting a problem, but weak as a standalone measure of typical play.
Use minimum FPS only with context. A frametime plot shows whether the low point was isolated or part of repeated instability, while 1% Low describes a broader slice of the slowest frames.
A fair handheld benchmark needs context
Handheld performance changes with power, temperature and software configuration.
Before comparing results, check:
• Handheld model and APU
• Power profile or measured TDP
• AC or battery operation
• Display resolution and refresh rate
• Frame cap and V-Sync or VRR state
• Graphics preset and individual settings
• Upscaler, quality mode and render scale
• Frame generation status
• Driver and game version when materially relevant
• VRAM allocation or UMA frame-buffer setting
• Test route, duration and capture tool
Do not compare a frame-generation result directly with native rendered FPS unless both are clearly labelled. Generated frames can improve displayed fluidity, but they do not remove the need to report the underlying rendering conditions.
Temperature also matters. A short run from a cold start may not represent sustained handheld performance. Allow the device to warm up, repeat the same route and use more than one run when possible.
A repeatable test method
For a useful community benchmark:
• Record the exact device, power profile, resolution and graphics settings.
• Warm up the game and device before capture.
• Use the same save, route or built-in benchmark for every comparison.
• Keep the capture duration consistent.
• Repeat the run and investigate any result that differs sharply.
• Save the raw capture, frametime graph or an uncropped screenshot that identifies the test.
• Report the capture tool and distinguish measured data from personal impressions.
NVIDIA notes that FrameView records detailed per-run data and can hide its overlay during capture to reduce measurement interference. AMD describes OCAT as a low-overhead capture and analysis tool built for frame-pacing data. Whichever tool you choose, consistency matters more than mixing results from different methods.
How to read a HandheldAtlas result
Read a benchmark in this order:
• Frametime evidence: Is delivery stable, and are spikes isolated or repeated?
• 1% Low: If reported, how close is it to the average?
• Average FPS: Does it meet the intended 30, 40, 45 or 60 FPS target?
• Test conditions: Are TDP, resolution, upscaling and frame generation clearly stated?
• Proof: Does the source link open the actual screenshot, capture or test—not a logo or unrelated image?
A trustworthy preset does not need every metric, but every number it includes must be supported. Unknown values stay blank. That is more useful than a complete-looking table built on guesses.
The practical takeaway
Average FPS tells you how much work was completed. Frametime shows when each frame arrived. 1% Low summarizes the slowest meaningful portion of the run. Together they distinguish a fast benchmark from a smooth gaming experience.
For handhelds, the most credible result is not necessarily the one with the largest FPS number. It is the one with repeatable conditions, stable frame delivery and evidence you can inspect.
Primary sources
Primary documentation used for this guide:
• NVIDIA FrameView — https://www.nvidia.com/en-us/geforce/technologies/frameview/
• NVIDIA FrameView 1.9 User Guide — https://images.nvidia.com/content/geforce/technologies/frameview/frameview-1-9-user-guide-web-version.pdf
• Intel Unreal Engine Optimization and Profiling Fundamentals — https://www.intel.com/content/www/us/en/developer/articles/technical/unreal-engine-optimization-profiling-fundamentals.html
• AMD GPUOpen OCAT — https://gpuopen.com/ocat/
