CPU: one plug-in against a desktop stack, measured on a Mac
Measured 8 November 2026 on an Apple M4 Pro Mac. These are Mac numbers, not iPad numbers.
We often say a full OneMove Vocal Console chain is light. On iPad we measured it at about 6 % of one core (iPad Air M3, method, measured before the Space module was added). But "light" only means something next to the alternative: a chain of separate desktop plug-ins doing the same jobs. So we built those chains from plug-ins installed on a Mac and measured both sides the same way.
The short answer
| chain | CPU (one Mac core) | plug-in latency |
|---|---|---|
| OneMove Vocal Console, all 14 modules, Studio, classic noise reduction | 6.0 % | 32 ms |
| OneMove Vocal Console, all 14 modules, Studio, neural noise reduction | 10.4 % (5.8 % on the audio thread) | 62 ms |
| OneMove Vocal Console, all 14 modules, Live | 5.5 % | 4.0 ms |
| Desktop stack, 9 plug-ins, light reverb | 8.7 % | 105 ms |
| Desktop stack, 9 plug-ins, plate-emulation reverb | 16.6 % | 105 ms |
| Desktop stack, 16 plug-ins, one per OneMove module | 22.0 % | 108 ms |
So on this Mac, the desktop stacks used 1.5 to 3.7 times the CPU of our chain with classic noise reduction. With our neural noise reduction switched on, the gap shrinks: the light 9-plug-in stack used less than we did (8.7 % against 10.4 %), and the larger stacks used 1.6 and 2.1 times as much.
The biggest single factor was the reverb. One plate-emulation reverb cost 7.7 % on its own, more than our whole chain with classic noise reduction.
What's in the desktop stacks
All plug-ins were already installed and licensed on the Mac. We used factory vocal presets wherever the plug-in has one, so anyone with the same plug-ins can rebuild the chains.
| job | 9-plug-in stack | 16-plug-in stack adds |
|---|---|---|
| noise reduction | Waves Clarity Vx, preset Reduce Constant Noise | |
| pitch correction | Antares AutoTune Advanced, HQ, Retune 19 | |
| dynamic EQ | Waves F6, Vocal Cleaner | |
| EQ | Waves H-EQ, Vocal 1 | |
| opto compressor | Waves CLA-2A, Vox | |
| FET compressor | Waves CLA-76, Vocal | |
| saturation | Waves Kramer Tape | Waves Magma BB Tubes, Vocal Warm Drive |
| reverb | Waves Abbey Road Plates, Vocal Plate (or Waves RVerb, Big Plate, for the light stack) | Waves Doubler2, Vocal Doubler |
| limiter | Waves L2 | |
| more | Waves C1 gate, Vocal Rider, Vocal Bender, OneKnob Louder; Slate Digital Fresh Air |
Each plug-in alone, on the same Mac: the plate reverb 7.7 %, the tube saturator 2.7 %, Clarity Vx 2.4 %, the tape 2.0 %, H-EQ 1.2 %, AutoTune Advanced 1.1 %, F6 0.8 %, CLA-2A 0.6 %, and the rest 0.5 % or less. The single figures add up to the chain totals, so chaining them costs nothing extra.
How we measured it
- One host for everything. A small test host loads each installed Audio Unit, chains them, and renders the same 20-second synthetic singer through them (the same signal as our iPad benchmark), at 48 kHz in 256-sample blocks.
- The plug-ins think they are live. We render in real-time mode, so no plug-in switches to an offline path. Offline mode gave the same numbers within half a point.
- What we count. The CPU time of the whole process, including every plug-in's helper threads (Clarity Vx's neural model and ours run on their own threads), divided by the length of the audio. That's percent of one core.
- Fair repeats. Every run starts a fresh process, and the chains take turns, five rounds each, so changes in what else the Mac is doing hit all of them alike. We report the median; the spread was under half a point.
- The same kind of core. Apple chips have fast and efficient cores. We checked that 95–100 % of the work ran on the fast ones in every run.
What this doesn't show
- It's not an iPad measurement. Our own chain costs about the same on this Mac core as on an iPad Air M3 (6.0 % against 5.7 %), but desktop plug-ins don't run on iPad, and iPad plug-ins are different products that each run in their own process. A stack of iPad plug-ins may cost more or less than these.
- Settings change everything. A convolution reverb or heavier presets cost more, lighter ones less. That's why we show three stacks and name every plug-in.
- CPU isn't quality. These plug-ins do things ours don't (see the full comparison).
- Latency adds up too. The desktop stacks report about 105 ms, mostly the neural denoiser (43 ms) and the tuner in its HQ mode (58 ms). Our Studio mode reports 32 or 62 ms, and Live mode 1.3–4 ms.
Waves, Clarity Vx, CLA-2A, CLA-76, F6, H-EQ, RVerb, L2, C1, Vocal Rider, Vocal Bender, OneKnob, Doubler and Magma are trademarks of Waves Audio Ltd. Antares and Auto-Tune are trademarks of Antares Audio Technologies. Fresh Air and Slate Digital are trademarks of Slate Digital. Abbey Road, Kramer and other names are trademarks of their owners. Products are named only for comparison; OneMove is not affiliated with or endorsed by any of these companies. Measured 8 November 2026 with the versions installed on that day.