Measurements of 1080p camera performance

8 September 2026 · Apple M2 · 8 logical CPUs · 24 GiB memory · Darwin 23.6.0

Measurement data (JSON) · Artifact manifest · SHA-256 checksums

Configuration and method

This research fixture uses task-local ordered export and a bounded queue of raw frames. The queue processes frames in arrival order. The fixture includes experimental changes that the released product does not contain. These results do not establish performance of the released product or physical display output.

Resolution and source	Noisedeck source A renders a GLSL test pattern at 1920 × 1080. The pattern contains sequence and time markers.
Data path	Ordered export sends frames to the Sync camera source. Receiver B reads the borrowed NV12 input. The ordinary media effect renders the final canvas. The observer reads that canvas after the same render.
Web runtime	Chrome 152.0.7977.77 runs the source and receiver in separate processes.
Desktop runtime	Electron 42.3.0 / Chrome 148.0.7778.180 runs the source and receiver in separate processes.
Run continuity	One source and receiver remain active for each run. The stream does not reset between report intervals.
Pixel controls	Three positive controls and three poison-negative controls check all pixels for each context before timing. During timing, the observer records marker identity. It does not archive full-color frames.
Selection and timing	The observer selects pairs with exact upload ordinals. It uses one-second intervals on a common local clock. Each interval includes its start and excludes its end. Late completions retain their reserved identities. Unique FPS divides unique pairs by the actual measurement duration, including the partial tail.
Acceptance criteria	Average ≥50 unique frames/s, every complete second ≥50, and sustained 60 are separate criteria.

Measurement results

All three runs passed the diagnostic and cleanup checks. All selected input/final pairs matched, with zero invalid or reordered identities. Each run passed the average-50 criterion. No run passed the every-second-50 criterion or the sustained-60 goal. The report counts duplicate pairs separately and excludes them from unique FPS.

Run	Actual duration (s)	Selected pairs	Unique	Duplicates	Unique FPS	Seconds <50 / complete seconds	Minimum
Web · 3 hours	10800.017199999	642,082	628,778	13,304	58.220092464	787 / 10,800	0
Web · 30 minutes	1800.020400000	107,569	105,882	1,687	58.822666676	92 / 1,800	41
Desktop · 30 minutes	1800.008700001	105,932	105,877	55	58.820271258	68 / 1,800	43

Partial final cells

Run	Duration (ms)	Input pairs	Final pairs	Final unique
Web · 3 hours	17.199998856	1	1	1
Web · 30 minutes	20.399999619	1	2	2
Desktop · 30 minutes	8.700000763	1	1	1

Three-hour one-second unique counts and a detail view of minutes 73 through 83. — Figure 1. Blue lines indicate unique counts in complete one-second cells. Dark horizontal lines indicate five-minute means. Reference lines indicate 50 and 60. Both panels use the same scale, which starts at zero. Input/final counts can cross a cell boundary, but totals for the full measurement match. SVG

Intervals with low counts in the three-hour run

The three-hour run contains 787 complete seconds below 50 unique frames/s and one zero-pair second. At measurement second 4,928, approximately 14:18:28.720 UTC, both input and final observers recorded zero admitted pairs. The following second recorded 13 pairs: 9 unique and 4 duplicate. This is an observed-upload gap, not an invalid-marker event.

A receiver snapshot by minute 90 recorded a maximum render gap of 1,669.8 ms. The record does not contain its exact event clock or source progress for each second. Four new receiver processes first appeared in the adjacent resource sample. The times correlate, with approximately 55 ms of uncertainty in clock alignment. The cause remains unresolved.

All 11 complete seconds below 30 unique frames/s

Second index	Elapsed mm:ss	Pairs	Unique	Duplicates
4445	74:05	35	25	10
4463	74:23	26	19	7
4600	76:40	36	26	10
4881	81:21	29	29	0
4882	81:22	19	18	1
4883	81:23	23	19	4
4884	81:24	20	18	2
4885	81:25	27	24	3
4927	82:07	26	25	1
4928	82:08	0	0	0
4929	82:09	13	9	4

Non-overlapping five-minute means for each run. — Figure 2. The plot uses means from complete five-minute intervals that do not overlap. Its vertical range is 54–61. These means exclude partial tails. SVG

The final five minutes contain 17,521 unique pairs and 316 duplicate pairs. The mean is 58.403333 unique frames/s. The minimum is 42, with 18 complete seconds below 50. The interval series does not indicate monotonic degradation across the full run.

Three-hour five-minute intervals

Minutes	Unique	Duplicates	Unique FPS	Minimum one-second count	Seconds <50
0–5	17,659	268	58.863333	41	15
5–10	17,644	282	58.813333	42	17
10–15	17,566	342	58.553333	41	22
15–20	17,671	261	58.903333	42	15
20–25	17,663	259	58.876667	43	14
25–30	17,548	342	58.493333	40	19
30–35	17,688	239	58.960000	42	14
35–40	17,663	255	58.876667	41	12
40–45	17,559	341	58.530000	40	19
45–50	17,680	227	58.933333	39	8
50–55	17,543	401	58.476667	41	19
55–60	17,499	417	58.330000	42	19
60–65	17,544	370	58.480000	43	15
65–70	17,649	283	58.830000	41	13
70–75	16,485	726	54.950000	19	61
75–80	16,886	496	56.286667	26	45
80–85	16,867	467	56.223333	0	40
85–90	17,570	300	58.566667	42	14
90–95	17,428	421	58.093333	43	27
95–100	17,408	460	58.026667	42	24
100–105	17,445	440	58.150000	40	25
105–110	17,563	323	58.543333	45	19
110–115	17,498	370	58.326667	41	21
115–120	17,434	416	58.113333	42	22
120–125	17,440	391	58.133333	42	21
125–130	17,527	331	58.423333	42	18
130–135	17,440	404	58.133333	41	23
135–140	17,379	477	57.930000	41	27
140–145	17,398	447	57.993333	42	27
145–150	17,567	336	58.556667	42	19
150–155	17,456	364	58.186667	41	19
155–160	17,360	477	57.866667	41	31
160–165	17,438	387	58.126667	42	28
165–170	17,613	265	58.710000	42	14
170–175	17,478	403	58.260000	43	23
175–180	17,521	316	58.403333	42	18

Web and desktop thirty-minute one-second unique counts. — Figure 3. Each web and desktop run contains 1,800 complete seconds. Both panels use the same scale, which starts at zero. Similar averages from separate runs do not establish an effect from the browser version. SVG

Resource sampling results

The sampler reads process counters approximately every ten seconds. CPU percentages divide the actual change in cumulative CPU counters by the actual interval for each process. A value of 100% equals one core. RSS and physical footprint are separate sums of process memory stocks. They do not measure uniquely owned memory, swap or GPU memory.

Web segment	Sample entry times (UTC)	Samples	Original result	Gap from prior final entry
Segment 1	13:20:09.732–14:18:29.765	351	Failed endpoint. Valid sample prefix	—
Segment 2	14:29:10.489–14:57:10.486	169	Complete	640.724 s
Segment 3	14:58:13.751–15:26:13.766	169	Complete	63.265 s
Segment 4	15:29:06.310–15:56:16.334	164	Failed endpoint. Valid sample prefix	172.544 s

Only segments 2 and 3 completed their declared 28-minute spans. Segments 1 and 4 retain valid sample prefixes and failed endpoints. Resource samples do not cover approximately the first 24 minutes. Later gaps remain separate. Resource samples do not cover the full three-hour run.

Four separate memory sampling segments for source, receiver and native publisher. — Figure 4. Each column contains a separate span of samples from groups of web processes. Solid lines indicate physical footprint. Dashed lines indicate RSS. The JSON retains the number of processes in each group. No line crosses a gap between spans. SVG

In segment 1, the receiver group’s footprint increased by approximately 252 MiB. The original processes contributed 151 MiB. Four new sampled processes contributed 101 MiB. The next two complete spans changed less. These observations do not establish a memory leak.

CPU counter intervals for selected processes, separated by sampling segment. — Figure 5. CPU intervals for four web processes: source renderer, receiver helper, receiver renderer and native publisher. Each process has a separate plot. The name “receiver helper” does not identify a GPU process. SVG

The native publisher’s CPU means were approximately 29.3%, 27.5%, 27.8% and 27.9% of one core across segments 1–4. These means use the actual interval durations as weights. Segment 4 failed while target processes stopped. It does not count as a completed observation.

Figure 6. The desktop sample contains one completed 28-minute span inside the 30-minute run. It contains 169 samples from a fixed group of 15 processes. The desktop plots sum CPU within each group. The web plots contain separate processes. Resource samples cover only part of the camera run. SVG

The desktop receiver group’s footprint increased by 44.28 MiB during the sampled span. The source group’s footprint decreased by 9.14 MiB. An increase in a memory stock or a decrease in RSS alone does not establish leak or swap behavior.

Limitations

These are separate single-host research runs. The tests do not include a controlled comparison with released builds. The report does not estimate statistical uncertainty.
The measurements include observer and sampler overhead. They do not subtract an estimate of that overhead. The investigation recorded host metadata. Browser user-agent compatibility strings do not identify the hardware.
Marker identity checks the selected input/final association during timing. It does not establish full-color pixel equality for every timed frame or physical display delivery.
The sampler misses processes that start and exit between samples. The report retains failed endpoints and gaps in coverage. Resource changes near a delivery gap do not establish its cause.

Source and checksum references

The JSON contains every complete one-second count, partial tails, five-minute summaries and separate resource series. A separate archive retains the original receipts and sources. Public data excludes local paths, process identifiers and private configuration.

Evidence SHA-256 values

Web · 3 hours receipt	`a5eab0112d693be578a9869a87fa445db856334fdeef23d309472101a71b2ce4`
Web · 3 hours cell journal	`62a388862a9b5a28040dc02f0d8deb1805bb36463e21143b66b15886881f6151`
Web · 30 minutes receipt	`0d5d92b585910ea5e9b63c2641cea75dd31408037c1df6c0dad39dd0a8c8f671`
Web · 30 minutes cell journal	`e93b0432456a3f38ee036f60cef3a203776bcc829c4dc2753f8e4a9e0859fa2e`
Desktop · 30 minutes receipt	`3db7da650536161a63805d9d5859f1ec16f401f2000aecb519e27e6c42197ac0`
Desktop · 30 minutes cell journal	`4ab42a3d5af56a3c0f7bca60412eb5cfb425eb398ad6355a6651b64062f57f49`
Web resources Segment 1 receipt	`68f814736e23a7794b6e2332f269598393d3a27219c95aea4cf309798a01a66f`
Web resources Segment 2 receipt	`a3a53eeda320107efd4558ae32ab58b4f2d67867a7f992f211dd012ba40c4c37`
Web resources Segment 3 receipt	`6d057592801165b0d766b09541d850a0b661c5f112449a13f7707738603f4878`
Web resources Segment 4 receipt	`2ff8b45dfd47ce1e2449a6989fb712b142445a98396ee78b4041a0665f81fb5e`
Desktop resource receipt	`f35cbd082df440ee0f35c026d8030437e2b6ddd4260792909eb0a934bd76bc9d`
Research fixture manifest	`2c96ae157ae78d272f88f806bd3907823543273b6da0a67bd03937951408dfc6`
Research native executable	`b960a686d44eeeba51a8b519c875ad955bffbc0b32e2cccf1e1c4e299b6aa344`
Archived source manifest	`c8fbc0fbadf56459012adb964bbe53a7efeab98285e8a7c31c3a8e371850d8c0`
Independent FPS audit	`fe30b0d9e1f3acba55920aa86c7e4ccf3d478353a258d8ffd9faa2b1b38ecb63`
Independent resource audit	`984168b745fca20f2a9b37b156c8f37a1181645ed448779eaba6d8df101cd06f`
SDK served-byte proof	`8b4f22609d63bdbaf67e0e1e2c732ac5a7f43583103483b7ccfacd8ae1699f21`
Sync 0.2.57 release JSON	`4c33d55a99f610a68df70efc473d903addfef13e005248e115901f7472338b6e`
Release and guide public proof	`01250a785bcd335d8db0a38d11dd1fe43e18ff9da15bc427db9bd19238a8fbfa`

Artifact ID: research-2026-09-08-1080p60-camera. Manifest · Checksums