You open a single pad, hold a chord, and the meter shoots to 70 percent. Sound familiar? You’ve got company. Since the update dropped, high Serum 2 CPU usage is the complaint I hear more than any other, and honestly, it’s justified.
Building Serum presets is my day job at Monosounds. No patch ships in one of our packs until it passes a CPU test on a mid-range laptop. Which means I’ve sunk an absurd amount of time into slimming down heavy patches without wrecking their sound.
Below are 12 fixes, in ranked order. The first four handle most cases. The final one deals with an odd GPU bug that masquerades as a CPU problem.
Why Serum 2 CPU Usage Runs Higher Than Serum 1
Straight talk first: on certain patches, Serum 2 really is heavier than Serum 1. Not universally. A basic wavetable patch with reasonable settings costs roughly what it cost back in 2020.
The new features stack up quickly though. There are now 3 oscillators rather than 2, and each can switch between 5 engines. Granular runs as many as 256 simultaneous grains. Spectral resynthesis reconstructs sound partial by partial, live. The FX section grew from a single chain into multiple buses with modulation per effect. That’s dramatically more DSP than the old engine ever needed. The full comparison is in Serum 1 vs Serum 2.
Xfer is aware of all this. Official CPU guidelines live on their support page, and the fixes here follow that same logic, plus everything I’ve picked up building presets on the new engine since launch day.
The Big Four: Voices Are Everything
1. Cut Unison to 3-7 Voices
Unison comes first because voices multiply everything else. At unison 16, the oscillator, the filter and every mod source run 16 times per note. Hold a 4-note chord and you’re already at 64 voices before a single effect kicks in.
Take it down: 3 for bass, 5 for leads, 7 tops for massive supersaws. Set detune around 0.10 to 0.15 and push Width, and unison 5 gets you roughly 90 percent of the spread of 16 for a third of the price. Barely any of the bass presets we sell go above 3.
2. Fake the Width With Chorus Instead
The trade almost everyone overlooks: chorus costs less than unison. A chorus sitting on an FX bus processes the summed signal exactly once, regardless of how many notes and voices are sounding.
So set unison to 3 and let a chorus supply the smear and stereo spread. On our supersaw patches this single swap dropped CPU by 40 to 50 percent, and in a blind test nobody on the team could consistently identify the unison 16 version.
3. Cap Polyphony
Match Serum’s poly limit to what the part genuinely uses. Bass: 1 or 2 voices. Keys: 6 to 8. Pads: maybe 8 to 12. The factory limit sits way above what most parts ever need, and every voice you allow is a voice that can accumulate.
Note that released notes still count toward polyphony — which leads directly to fix four.
4. Shorten Release Times
A note keeps burning CPU after you let go, right up until the amp envelope closes completely. With an 8-second release, every note you play lives on for 8 extra seconds. A few chord changes later, there are 30 invisible voices humming away behind the music.
Bring the amp release under 2 seconds and hand the tail to a reverb. A reverb’s cost is fixed. Voice tails grow with everything you play.
Engine and Quality Settings
5. Oversampling: 2x, Not 4x
Oversampling lifts the internal processing rate to fight aliasing, and CPU cost climbs right along with it. 2x is where the value is. 4x roughly doubles the bill again for a difference that mostly surfaces on bright, high-pitched patches heard in isolation.
My policy: work at 2x, and when a lead gets gritty in the upper register, render just that part at 4x. Not once has anyone emailed me about aliasing in a finished track.
6. Watch the Granular Grain Count
Up to 256 grains can run in the granular engine, and each grain is a miniature playback voice carrying its own envelope. Put 256 grains under a unison stack and you’ve lit a CPU fire.
Most granular textures sound identical with grain count and density kept sensible. And when you go granular, leave unison at 1 — the engine hands you thickness for free already.
7. Bounce Spectral Patches When You’re Done
Spectral is Serum 2’s most expensive engine. Resynthesizing hundreds of partials in real time is fantastic for sound design and brutal for playback.
When the sound is where you want it, hit Resample to Oscillator to print it into a wavetable, or just bounce the track to audio. The sound stays; the cost collapses.
8. Turn Off What You’re Not Using
Every module that’s switched on costs something, even when its effect is barely audible. Second filter doing almost nothing? Kill it. Third oscillator at 5 percent in the mix? Mute it and check whether you notice. An effect running at 3 percent mix? Gone.
This unglamorous sweep reliably trims 10 to 20 percent from bloated patches. It’s step one whenever I dissect presets, and skipping it is among the habits I flag in Serum 2 beginner mistakes.
Project-Level Fixes
9. Simplify the FX Buses
Multiple FX buses are wonderful until every bus is hauling its own reverb. Reverbs and the Bode frequency shifter sit near the top of the price list, and per-effect modulation stacks more on top of that.
One reverb per patch, maximum. And when three patches in a project all want the same big hall, a single reverb on a DAW send beats a reverb inside every Serum instance.
10. Raise Your Buffer Size
Completely free, this one. Track at 128 or 256 samples so latency stays playable. Come mix time, lift the buffer to 512 or 1024. Per-patch Serum 2 CPU usage doesn’t move, but the crackling stops because the computer gets more time to finish each block.
11. Freeze and Bounce
The oldest trick remains undefeated. When a Serum part is locked, freeze the track or print it to audio. An audio file costs essentially nothing, however monstrous the patch behind it was.
Anything that goes two sessions without an edit gets bounced, and I stash the MIDI plus the preset name in the track notes so the road back always exists.
12. The Serum.cfg Fix for GPU Glitches
Some “CPU” problems are graphics problems in disguise. Interface stuttering, meter spikes that only appear with the plugin window open, glitchy visuals — try switching OpenGL off.
In Serum 2, head to the Menu tab and pick Open Serum 2 Presets folder. From there, enter the System folder and open Serum.cfg in a text editor: set UseOpenGL = 0, save, restart your DAW. On machines running flaky GPU drivers this eliminates spikes that no synth setting could ever reach.
And should the plugin disappear from your plugin list after system tweaks like this one, here’s how to fix Serum 2 not showing up in your DAW.
Heavy vs Lean: The Cheat Sheet
Here’s the checklist I run whenever a patch flunks our CPU test:
| Setting | Heavy patch | Lean patch |
|---|---|---|
| Unison | 16 voices | 3-7 voices |
| Width source | Stacked unison | Chorus on an FX bus |
| Polyphony | Default, uncapped | Matched to the part (1-12) |
| Amp release | 6-8 seconds | Under 2 s, reverb carries the tail |
| Oversampling | 4x everywhere | 2x, render 4x if needed |
| Grains | 200+ | Modest count, unison 1 |
| Reverbs | One per FX bus | One total, or a DAW send |
Look closely: none of these moves really touch the character of the sound. CPU lives in voices, tails and oversampling. Character lives in the wavetable, the filter and your FX choices.
It’s precisely how we work at Monosounds. Every patch across our Serum 2 preset bundles has to run clean on a 2019 laptop, and the approach is free to inspect: load a few patches from our 250+ free Serum 2 presets and look at the unison and poly settings yourself.
Serum 2 CPU Usage: Common Questions
Is Serum 2 really heavier than Serum 1?
For some patches, yes. Anything running granular, spectral or several FX buses is simply doing more math, and math costs. Straight wavetable patches land close to Serum 1 territory. Given that the upgrade is free for Serum 1 owners, the CPU tax is really the only entry fee.
Does 2x oversampling sound worse than 4x?
Inside a full mix, you will almost never tell them apart. Aliasing appears on bright, high-pitched material played solo. When one specific lead bugs you, bounce that single track at 4x and keep the rest at 2x.
Why does one patch use 40 percent CPU on its own?
Voices, nearly every time: unison multiplied by polyphony multiplied by release time. Audit those three before you blame the synth. A unison 16 patch carrying a 6-second release can quietly stack 80+ voices out of an ordinary chord progression.
Will a faster computer fix Serum 2 CPU usage?
It lifts the ceiling, sure, but voice discipline scales better than any hardware. I’ve seen a capped, tidy project push 40 tracks on an aging laptop while a sloppy 12-track session strangled a brand-new machine. Repair the patches first; upgrade only if you still have to.
Start With Unison, End With Bounce
Go down the list from the top. Unison, polyphony, release and oversampling clear 80 percent of cases in around five minutes. Freezing covers most of what remains, and Serum.cfg covers the machines where CPU was never actually the problem.
Prefer patches that already pass the test? Grab our 250+ free Serum 2 presets. No strings attached. Pull them apart, audit the voice counts, and steal any trick you find in there.
I run Monosounds.studio and make the packs myself. Ten-plus years of producing, 97+ releases, 16,000+ downloads. Every Serum 2 preset we ship passes a CPU check on a mid-range laptop and gets used in a real track before release. If it does not survive a mix, it does not go in the pack.
More about Maxim & the studio →Looking for the packs themselves? Here is the whole Serum 2 catalogue — sorted by genre and by sound, with a demo on every one.
