One prototype, one machine pair.
Latency measurements
These are software timestamps from the prototype's own instrumentation, not an independent benchmark. They are published so the figure quoted elsewhere on this site can be checked rather than taken on trust.
What is measured
The interval from the Windows capture callback — the moment a new frame is handed to DualOS — to the point where the decoded frame is enqueued into the macOS display layer. It covers window capture, hardware encoding on the Windows GPU, QUIC transport, decoding on the Mac, and submission for presentation.
It does not cover macOS compositor timing, display scanout or panel response. This is not a glass-to-glass figure. What reaches your eye is higher by whatever those three stages cost on your hardware.
Results
| Native | Half resolution | |
|---|---|---|
| Scene | 3456×2144 | 1748×1028 |
| Codec | HEVC 10-bit 4:4:4 HDR10 | HEVC 10-bit 4:4:4 HDR10 |
| Bitrate | 16.1 Mb/s | 3.7 Mb/s |
| Median (p50) | 9.8 ms | 7.9 ms |
| p95 | 17.4 ms | 12.1 ms |
| p99 | 21.1 ms | 17.5 ms |
| Max | 26.6 ms | 34.4 ms |
| Jitter | 7.6 ms | 4.3 ms |
| Encode (Windows) | 4.3 ms | 4.7 ms |
| Network | 1.2 ms | 0.8 ms |
| Decode (Mac) | 3.5 ms | 2.3 ms |
| Round-trip time | 1.0 ms | 0.5 ms |
| Clock sync | ±0.23 ms | ±0.37 ms |
| Dropped frames | 19 | 0 |
| Sample | 1047 frames / 10 s | 351 frames / 10 s |
At 120 Hz one frame budget is 8.33 ms. The native-resolution median sits just over one budget; the 95th percentile sits at two.
Conditions
Windows 11 with an RTX 5080, an Apple Silicon Mac with a 120 Hz display, wired local network.
Frame rates in these samples are not representative of throughput. Windows Graphics Capture delivers frames on content change, so a still window produces few frames no matter what the pipeline can carry.
What will move these numbers
These numbers vary with the GPU used for encoding, the Mac chip used for decoding, the network connection, resolution, chroma and bit depth. Nothing here has been measured on another machine.