u-he Diva CPU Guide: How to Run Diva Without Overloading Your Computer
What Diva's accuracy modes, OfflineAcc, Multicore, Voices and Stack actually do to your CPU, plus a simple workflow to keep sessions stable without losing the sound.
What Diva's accuracy modes, OfflineAcc, Multicore, Voices and Stack actually do to your CPU, plus a simple workflow to keep sessions stable without losing the sound.

Diva sounds great and it knows it: load a few pads on the top quality setting and your CPU meter goes red. The good news is that Diva CPU usage is very controllable once you know which settings actually matter. After this guide you'll know what each accuracy mode does, when Multicore helps and when it hurts, and a simple workflow that keeps sessions smooth without giving up the sound.
u-he describes Diva as a realtime circuit simulation with zero delay feedback filters. In plain words: instead of playing back a clever approximation of an analogue synth, Diva calculates how the circuit would behave, sample by sample, for every voice you play. That's where the warmth and the way the filter reacts to resonance come from, and it's also why a four note chord costs four times as much as a single note.
So the total load comes from three things multiplied together: the accuracy setting, the modules you choose, and how many voices are sounding at once. Fix any one of them and you'll feel it.
The Accuracy selector sits in the Main panel. These are the four modes and how the Diva user guide describes them:
| Accuracy | What u-he says | When I use it |
|---|---|---|
| draft | CPU friendly, but FM will sound rough and resonance is primitive | Old laptops, sketching chords, never for final sound design |
| fast | Fine for older computers or when you need more polyphony | Writing sessions on a weaker machine |
| great | The best compromise between quality and polyphony | My default for writing and mixing |
| divine | Top quality zero delay feedback filters, seldom necessary | Exposed solo sounds, and only when the CPU allows it |
Two details people miss:
One warning: because FM and resonance change audibly on draft, don't design a sound on draft and then judge it again on great. Pick the setting you'll write with first, then tweak.
Multicore is also in the Main panel. It distributes the voices of one Diva instance across your CPU cores so you can play more notes. Sounds like an easy win, but the manual adds an honest caveat: Multicore can even reduce performance if your CPU is older or much newer, and it names Apple Silicon (M1, M2 and later) as an example.
My approach is to test it rather than assume:
Keep in mind that your DAW already spreads separate tracks across cores. Multicore inside Diva mainly helps when one single instance plays lots of voices, like a wide pad with Stack. For a mono bass it does very little.
In the voice settings, Voices sets the maximum number of voices, from 2 to 16, before note stealing kicks in. The manual says it's mostly used to guard against audio glitches with CPU intensive presets. If a pad is choking your system, dropping Voices from 16 to 6 often fixes it with no audible difference, because you rarely hold more than five or six notes including release tails.
Stack is unison: each note you play becomes several voices. The manual puts it bluntly: stacking will eat up CPU. Stack 4 on a four note chord is sixteen voices. If you want width, try two stacked voices with StackIndex panning (I show that in the StackIndex dual filter trick) instead of four or more.
Also check the Mode. Basses and leads usually work in mono or legato, which means one voice. For pads, poly2 steals voices that are in their release stage first, which can help long release pads stay under the voice limit.
The oscillator model changes CPU load a lot. The manual calls Triple VCO the most CPU hungry oscillator model and says Dual VCO Eco has the lowest CPU hit. DCO, Dual VCO and Digital sit in between. On the Digital oscillator, the High Quality button reduces aliasing but usually costs a bit more CPU.
In practice I keep Triple VCO for mono basses and leads, where one voice is playing anyway, and use Dual VCO or Eco for background pads and chords. If you want to know what each model sounds like before you swap, read Diva oscillator models explained.
FREE Glow Diva presets
23+ free Diva sounds from Glow to test these settings on: load a pad, change Voices and Accuracy, and hear what actually changes.
Get the free Glow presetsOutside Diva, your audio buffer is the next lever. While recording MIDI or playing parts in, I keep the buffer at 128 or 256 samples so the latency stays playable. Once the parts are written and I'm arranging or mixing, I go up to 512 or 1024. Latency doesn't matter much when nothing is being played live, and Diva gets far more room.
When a Diva part is finished, freeze it or bounce it to audio. Every major DAW has a freeze or bounce in place function. Keep the MIDI and the unfrozen instance in the project (or save the preset) so you can go back. With OfflineAcc on best, the freeze or bounce is rendered at top quality, so you actually gain sound quality compared to what you heard while writing.
u-he themselves call divine seldom necessary. Great with OfflineAcc on best gives you the top quality render without paying for it in real time.
Resonant and FM sounds fall apart on draft. If a preset sounds cheap, check the accuracy before you delete it.
On Apple Silicon in particular it can be slower. Measure it.
Stack multiplies every note. A wide pad with Stack 8 is the fastest way to crash a session.
Because it simulates analogue circuits in real time for every voice, including zero delay feedback filters. More voices, a higher accuracy setting and heavier modules all add up.
For most people: great for playing, with OfflineAcc set to best so exports render at top quality. Use fast on older machines.
Test it. u-he's manual warns that Multicore can reduce performance on Apple Silicon, so compare the CPU meter with it on and off.
No. Accuracy and OfflineAcc are global per instance and stay the same across presets.
With OfflineAcc on best, a freeze or offline bounce can sound slightly better than what you heard in real time, because it's rendered at top quality.
Once Diva runs smoothly you can stop worrying about the meter and focus on sounds. Try these settings on the free Glow Diva presets, and if you want a full library of ready to use sounds, the Complete u-he Diva Preset Bundle gives you 480+ presets for melodic house and techno. New to installing packs? Start with how to install u-he Diva presets.
Your cart is currently empty.
Start Shopping