The short version
| Tier | Oversampling equivalent | CPU vs that equivalent | Measured aliasing SNR | In-band flatness | Latency |
|---|---|---|---|---|---|
| Normal | none | n/a | none applied | not corrected, by design | zero |
| HQ | Custom AA, about the same as plain 16x | about 4x less | 64.8 dB | within 0.002 dB | about 114 samples |
| Ultra | Custom AA, about the same as plain 40x | about 7.6x less | 81.7 dB | within 0.0003 dB | the same as HQ |
- Aliasing SNR is how far the harmonics a clipper should make sit above the aliasing it shouldn't. Higher is cleaner. A plain hard clip with nothing done about aliasing measures about 12 dB on this test, which is why its aliasing is audible. Normal has no figure because it applies no anti-aliasing.
- CPU vs that equivalent compares each tier with reaching the same result by oversampling alone. Ultra is compared with plain 64x, the nearest practical setting to the 40x it matches.
- The test tone is deliberately harsh: 11 kHz driven 6 dB into the ceiling, so every harmonic the clipper makes is too high for the sample rate and folds back down. Flatness is measured from 20 Hz to 20 kHz.
All three tiers are in HardStop Basic. None of them is a paid feature.
What each tier does
Clipping creates harmonics higher than your sample rate can hold. They have nowhere to go, so they fold back down into the audible range as aliasing: grit with no musical relationship to the source.
Normal leaves this alone on purpose. It is a plain hard clip with an exact ceiling, zero latency, and the aliasing kept as character. Plenty of records were made on clippers that work exactly this way.
HQ and Ultra tackle aliasing in two ways. They run the clipper at a higher internal sample rate, so the harmonics have room before they are filtered out, and both add the same anti-aliasing of our own on top. Ultra runs at a higher internal sample rate than HQ.
What that saves in CPU
Oversampling alone is expensive: each doubling roughly doubles the work. Our anti-aliasing costs far less, and combining the two keeps these tiers practical in a real session.
In these measurements every path used the same oversampler and the same clipper, so the only difference is the anti-aliasing:
- Adding our anti-aliasing to an otherwise identical path raises the aliasing SNR from 54 dB to 82 dB. That is 28 dB cleaner for a little more CPU.
- HQ matches plain 16x oversampling at about a quarter of the CPU.
- Ultra matches roughly 40x oversampling. Getting there by oversampling alone, at the nearest practical 64x, would cost about 7.6 times the CPU Ultra uses.
Those ratios held across every condition tested, from 8 kHz to 11 kHz and from 6 dB to 12 dB of drive. In short, Ultra gives you what used to take 40x oversampling or more, for about an eighth of the processing.
Why they usually sound the same
On real music, the oversampled tiers converge. Measured on music, 4x and 8x land within about a decibel of each other. The dramatic differences in plugin shootouts come from test tones, which are built to provoke aliasing in a way music rarely does.
So if you switch from HQ to Ultra on a finished mix and hear no change, that is expected, not a fault.
When the difference is real
One combination separates the tiers: heavy drive on bright material.
With gentle clipping the tiers sound close. Drive hard into material with lots of high-frequency energy and the gap opens up quickly. That is the case covered in why clipping distorts on some sources and not others, and it is why HQ is built to hold up under heavy drive, not just to measure well under gentle clipping.
Latency
- Normal with the TP safety limiter off has zero latency. Not near zero: zero.
- HQ and Ultra have the same latency, about 114 samples (under 3 ms at 44.1 kHz). Your DAW compensates automatically, and because the two match, switching between them never shifts your timing.
- Turning the TP safety limiter on adds the limiter's fixed delay, and in Normal it also adds the same oversampling delay as HQ and Ultra, so with TP on all three tiers report the same latency. That amount stays fixed when you move the Lookahead knob, on purpose: latency that changed as you turned a knob would click.
Best settings by task
| Task | Setting | Why |
|---|---|---|
| Tracking or writing | Normal, TP safety limiter off | The only zero-latency setting. |
| Many instances in one session | Normal on tracks; HQ or Ultra on buses and the master | Per instance, HQ uses about 12 times Normal's CPU and Ultra about 27 times. |
| Mixing | HQ | Clean, latency compensated, nothing to think about. |
| Mastering | Ultra, TP safety limiter on | Ceiling at -1 dBTP as a safe middle ground, or -0.2 dBTP if you are pushing hard. |
| A specific character | Normal, drive up | On drums and distorted material the aliasing can be part of the sound. |
For mastering, aim for impact rather than a strict LUFS number: why your songs aren't loud on streaming platforms explains why chart tracks sit nearer -8 LUFS than the often-quoted -14. And the cleanest setting isn't always the right one.
When comparing tiers
With the TP safety limiter off, switching between Normal and HQ changes the plugin's latency, so your DAW re-compensates each time. That is expected. When comparing by ear, give the DAW a moment to settle and match the output levels, or you will be hearing level and timing differences instead of aliasing.
Not sure whether you are hearing aliasing or just too much clipping? The thirty-second test will tell you.