BENCHMARK · METHOD AND RESULT
How the 29 July 2026 benchmark was measured
A plain-language summary of one performance soak of the studio.live wall: what ran, on what, how it was scored, what came out, and what it does not show.
WHAT THIS IS
On 29 July 2026 we recorded one 180-second performance run of the studio.live wall with three videos playing at once. It is a baseline for comparing studio.live with itself over time. It is not a pass/fail test and it is not a hardware benchmark.
The figures on this page are copied unchanged from the written record of that run. The benchmark note on the home page refers to this run and to nothing more.
The 48 streams on the home page are a target for the stress test you can run on your own machine. This soak did not reach it: it played 3 streams.
WHAT WAS MEASURED
- Scene
- The wall at rest, with three video tiles loaded and all three playing for the whole run.
- Length
- 180 seconds (planned 180, recorded 180.0): 10,796 frames and 37 memory samples.
- Frame rate
- How many times per second the page’s animation callback (requestAnimationFrame) ran on the main thread, counted for each complete second. It is not the rate at which the browser presented frames to the screen.
- Frame time
- The gap between one animation callback and the next, summarized as percentiles and a histogram.
- Slow frames
- Frames longer than 33.4 ms and longer than 100 ms, and the long tasks the browser reports.
- Memory
- The JavaScript heap (usedJSHeapSize), sampled 37 times, and the browser’s own memory estimate at the start and the end of the run.
WHERE IT RAN
- Graphics
- NVIDIA GeForce RTX 5090 over Vulkan 1.4, through the ANGLE graphics layer.
- Browser
- Chromium-based browser, version 149.0.7827.55, in a visible window with a real-GPU profile.
- Display
- 1280 × 800 viewport at about 60 Hz. The refresh rate is an estimate from the animation cadence at rest, not a reading from the display.
- Build
- A local build of the wall player, not the live site.
- Version
- Build be1fef03.
- Video decode
- Software. Hardware decode was not confirmed.
PROTOCOL AND GATES
- Health check first
- Before the soak, a health check confirmed the window was drawing at full speed: 60.8 animation ticks per second at rest (a starved or background window runs below 20) and 33 ms per wheel step (a starved window takes over 300 ms). Verdict: healthy.
- Window state
- Visible and focused at the start, with 0 visibility changes and 0 hidden samples during the soak.
- Soak
- One 180-second recording of the wall at rest.
- Pass/fail
- None. The run asserts no frame-rate or memory threshold. It is a baseline, not a gate: its value is the shape of the frame-time tail and the memory slope.
RESULT
- Frame rate
- Over 180 complete seconds: minimum 59, 5th percentile 60, median 60, maximum 60 frames per second.
- Frame time
- p50, p90, p95, p99, p99.9 and maximum were all 16.7 ms. All 10,796 frames fell in the 16.7 ms bucket.
- Slow frames
- 0 frames over 33.4 ms, 0 frames over 100 ms, and 0 long tasks (0 ms in total).
- JavaScript heap
- 89.3 MB to 74.1 MB over 37 samples; least-squares slope −4.596 MB per minute.
- Browser memory estimate
- 181.8 MB to 165.5 MB; −5.407 MB per minute from the start to the end of the run.
- Errors
- 0 console errors, 0 page errors and 0 unexpected API failures.
WHAT IT DOES NOT SHOW
Software decode only. No number here is evidence about hardware video decode, and none should be quoted as a hardware result.
The frame rate is the main-thread animation-callback rate. It is not the rate at which frames reached the screen.
Animation callbacks fire on the display refresh, so frame times come in whole refresh intervals and about 16.7 ms is the smallest value this instrument can report. A flat 16.7 ms line means no missed refresh. It does not show low frame cost or spare headroom, and a cost increase that stays within the budget would not appear in it.
The JavaScript heap series counts JavaScript memory only: no page elements, no GPU-adjacent memory and no workers. It is weaker evidence than the browser’s own memory estimate, so the two are reported separately. Memory fell during these three minutes; three minutes cannot show what happens over a long session.
One machine, one browser version, one run and one scene of three videos. Other computers, other browsers, phones and larger walls were not measured here.
This page is the public summary of that recorded run.