Reply to thread

Message: [QUOTE="Bombastus, post: 92146, member: 2178"] Audient rates every Horizon interface at 0.99 ms round-trip latency, but only at 192 kHz with a 64-sample buffer. At 44.1 kHz, the same buffer produces a published 4.286 ms round-trip. Audient lists 3.938 ms at 48 kHz, 2.143 ms at 88.2 kHz, and 1.969 ms at 96 kHz. Those numbers change the practical reading of the [B][URL='https://goldmidi.com/community/threads/audient-has-introduced-horizon-thunderbolt-3-interfaces.77487/']Audient Horizon Thunderbolt 3 interface range[/URL][/B]. Sub-millisecond performance is real in Audient's published specification, but it is not the latency you should expect simply because a Horizon is connected to a Mac. If your sessions normally live at 44.1 or 48 kHz, the more relevant figure is about four milliseconds before your plug-in chain adds anything. It is still quick. It just is not 0.99 ms. [HEADING=2]The 0.99 ms figure needs its full context[/HEADING] Audient lists the same latency figures for the Horizon 8P, Horizon 16, and Horizon 32, all measured with a 64-sample buffer. Multiplying each published round-trip time by its sample rate produces roughly 189 to 190 samples across the entire table. The pattern stays remarkably steady from 44.1 through 192 kHz. The consistency matters because latency shrinks mainly as those samples occupy less real time at higher sample rates. Horizon does not suddenly enter a different published performance mode at 192 kHz. Moving from 48 to 192 kHz cuts the stated figure by 2.948 ms. In a conventional host round trip, a 64-sample input buffer and 64-sample output buffer account for 128 samples before additional path latency. Audient's totals leave roughly another 61 to 62 samples in the complete path. The company does not publicly split that remainder into converter, driver, safety-buffer, or mixer components. Audient also labels the table "DSP Mixer Roundtrip Latency" while describing Horizon elsewhere as a platform without proprietary DSP chips. The statements are not necessarily in conflict, but the wording deserves attention. A published interface figure is not a guaranteed DAW monitoring number once arbitrary plug-ins enter the path. [HEADING=2]Higher sample rates buy speed at a real cost[/HEADING] At 48 kHz, Horizon's stated 3.938 ms round trip is almost exactly double its 1.969 ms result at 96 kHz. Moving again to 192 kHz halves the figure to 0.99 ms. The relationship is unusually easy to see. The trade-off sits elsewhere in the system. A 192 kHz session processes four times as many samples per second as a 48 kHz session. Higher sample rates also create larger audio files and increase disk activity, while plug-ins have more work to do. A vocal tracking template with a few efficient plug-ins may tolerate 96 or 192 kHz comfortably on a fast Mac. A large production packed with oversampling synths, convolution reverbs, linear-phase processing, and heavy virtual instruments can hit the CPU wall much sooner. Low latency stops helping once the session clicks, pops, or overloads. Buffer size still matters too. Audient currently publishes the Horizon latency table only at 64 samples, so there is no first-party 32, 128, or 256-sample curve to quote. Exact Horizon round-trip figures at those settings need direct measurement rather than confident extrapolation. [HEADING=2]Native plugin monitoring depends on the whole signal path[/HEADING] The interface is only one part of what you hear while recording. DAW buffers, conversion, plug-in latency, software instruments, routing, and processing on the monitored channel or master bus all add to the path. A fast interface cannot erase delay created farther downstream. Some plug-ins are effectively cheap in latency terms. Others deliberately hold audio for lookahead, oversampling, linear-phase filtering, pitch processing, or spectral work. Virtual instruments can add another processing stage before sound even reaches the output buffer. Musical timing also does not have one universal latency cliff. [B][URL='https://doi.org/10.1007/s004260100075']Delayed auditory feedback in music performance[/URL][/B] shows timing disruption changing with delay and performance rate. A single millisecond figure cannot describe how every singer, drummer, guitarist, or keyboard player will experience monitoring. For a typical 48 kHz tracking session, Horizon's published 3.938 ms figure is the useful starting point. You still have to account for whatever the DAW and plug-in path adds. A clean tracking template can therefore feel tighter than a dense mix session on the same interface. Audient's table leaves 48 kHz at 3.938 ms and 96 kHz at 1.969 ms with the published 64-sample setting. If monitoring feels much slower than those baselines, extra delay is entering somewhere else in the session path. It gives you a practical place to start troubleshooting. [/QUOTE]

Name