Official Serum 2 thread!

[teilo]

Dune 3: 2%
Hive 2: 2%
Phase Plant (have to use 4x osc, because of 8x unison limit): 7%
Serum 1: 27%
Serum 2: 12%

Maybe Serum 2 added a smarter unison implementation?

The reason for the efficiency of Dune and Hive (and probably Phase Plant) is that they aren't actually playing multiple oscillators, but are using wave shaping techniques to achieve the same result, while only needing to play a single or very few waveforms. But that same technique cannot easily be used in other oscillator types, such as sample and granular.

[Touch The Universe]

"but are using wave shaping techniques to achieve the same result"

Can you elaborate ?

[tumface]

The reason for the efficiency of Dune and Hive (and probably Phase Plant) is that they aren't actually playing multiple oscillators, but are using wave shaping techniques to achieve the same result, while only needing to play a single or very few waveforms. But that same technique cannot easily be used in other oscillator types, such as sample and granular.

That's clearly not true, because Dune 3 and Hive 2 have the same level of efficiency when set to wavetable oscillator type. I'm not even sure how waveshaping would make unison oscillators work with any oscillator type, to be honest.

[Touch The Universe]

yeah, makes no sense

[Touch The Universe]

So it's kind of just wave-shaping like a Sawtooth wave to sound like it's being detuned is that right


ChatGPT said:
Yes — that's exactly the idea, and you're honing in on the core of what makes these synths efficient.

Here's the basic concept:

Instead of this:
You literally detune 8–16 separate sawtooth oscillators, each with their own phase and pitch offsets. This sounds great — thick, wide, and rich — but it's CPU-intensive, especially with polyphony.

They do this:
Plugins like Dune, Hive, and others often simulate that detuned sound using wave-shaping and clever unison algorithms. For example:

They might generate one sawtooth and then apply phase modulation, stereo spread, and detune simulation internally.

Some use interpolated waveform blending, where they morph the waveform shape over time or across a virtual unison spread.

Others use mathematical tricks to fake the sound of multiple oscillators by summing harmonics with controlled interference patterns.

So yes — it's as if they're just warping or shaping a sawtooth wave to behave like it's being detuned without actually playing multiple independent oscillators.

Bonus Analogy:
It's like a painter who makes it look like a complex texture with just a few brush strokes instead of layering 20 colors.

Let me know if you want to get into the technical DSP side of how they might simulate unison with low CPU — it's a fascinating area.

[SirKen]

I read on another forum that they have not tested it properly on AMD chips at all. Don't know how true that is.

That sounds made-up to me. It works fine on AMD.

Honestly, that just sounds like coping to me.

I'm able to use it without any significant problems. The plugin loads fast and works fine. I've reported some bugs and they've fixed all of them so far. I don't know how that's supposed to be coping.

And on my Mac, where I do most of my audio stuff, it runs even better than Serum 1.

There are some presets that are unusable, but I wasn't going to use those presets, anyway.

[SirKen]

Am I understanding things right that Intel implementation was more through than the AMD implementation? I wouldn't mind grabbing it at the intro price but I don't want to gamble on a future update that might fix the CPU issue, might being the key word.

Intel and AMD use the same instruction set architecture. In other words, they're compatible with each other and use the same machine code.

The differences lie in the different generations and models of the chips produced by both companies. Different hardware microarchitectures, which are compatible with the same code as each other, might have different characteristics. For example, on some microarchitectures (generations of chip) if you have instruction sequence XXX then YYY then ZZZ, it's slower than on a different microarchitecture. But then on that one, a different combination might be slower. And some microarchitectures only support some subset of extended instructions with extra capabilities.

This table shows a bunch of instructions and their latencies (latency in nanoseconds, as in data dependency chains, not audio latency) https://asmjit.com/asmgrid/ I sometimes use it when working on SIMD code.

By the way, I have no knowledge of the internals of Serum 2. My guesses are all just guesses. I could be totally wrong.

I read on another forum that they have not tested it properly on AMD chips at all. Don't know how true that is.

I wouldn't mind grabbing it at the intro price but I don't want to gamble on a future update that might fix the CPU issue, might being the key word.

Even on my old Threadripper PC, where Serum 2 performs significantly worse than Serum 1, Serum 2 is still totally usable. There are some presets that are unusable, but I wasn't going to use those presets, anyway. Serum is one of my favorite synths to *make* presets in, so that's what I'm usually doing.

Also, Serum 2 has no problem idling when it's not in use, so as long as you aren't using a bunch of instances all at the same time, no problem.

Honestly, that just sounds like coping to me. Especially if there really is a difference in performance between different chip vendors. They need to do better than that if they want to sell a synth for near 200 and without a license transfer.

lol, that sounds like forum garbage or someone thinking they are being clever and trying to strong arm Steve to lower the price.

Nope, it was posted by someone that is very active on the other two big music forums online. I need to check with him on where he got that info.

EDIT - I found a reference to that post here too

Hoping for a CPU optimization of the granulator, even on my M3 Max Mac, at the highest sample buffer in Logic 11.1.2 CPU maxes out with just a few voices played, this can't be right, any other granulator I use would laugh at this patch...

Anyone knows if Serum 2 runs worse on Apple Silicon vs Intel?

From the Gearspace thread:

Question:
I was just wondering if it was some kind thing of amd thing is all.

David Gamble replied:
Could well be. Most of the profiling work was on intel but I just got a 9950 myself so I can take a look at that too Could very well be the case that some cache size issues are responsible. I'm not super worried just yet

He answers many of the forum replies, even on reddit so go ask him. The synth has some of the best resampling options available and like someone else said there is no digital equivalent right now for filters or the spectrum oscillator. Lowering quality for old chip-sets and people who have bad hearing would be a dumb idea considering the M4, resampling and compatibility mode already exist.

That's a silly take for two reasons:
1. There are many synths with multiple quality modes like real-time, rendering etc... It is nothing new in the VST world
2. Ryzen CPUs from the last 2 generations are not old chip-sets.

I am trying to see if I am misinformed or whether there are any official updates and promises that I am not aware of.

[swilow11]

So it's kind of just wave-shaping like a Sawtooth wave to sound like it's being detuned is that right


ChatGPT said:
Yes — that's exactly the idea, and you're honing in on the core of what makes these synths efficient.

Here's the basic concept:

Instead of this:
You literally detune 8–16 separate sawtooth oscillators, each with their own phase and pitch offsets. This sounds great — thick, wide, and rich — but it's CPU-intensive, especially with polyphony.

They do this:
Plugins like Dune, Hive, and others often simulate that detuned sound using wave-shaping and clever unison algorithms. For example:

They might generate one sawtooth and then apply phase modulation, stereo spread, and detune simulation internally.

Some use interpolated waveform blending, where they morph the waveform shape over time or across a virtual unison spread.

Others use mathematical tricks to fake the sound of multiple oscillators by summing harmonics with controlled interference patterns.

So yes — it's as if they're just warping or shaping a sawtooth wave to behave like it's being detuned without actually playing multiple independent oscillators.

Bonus Analogy:
It's like a painter who makes it look like a complex texture with just a few brush strokes instead of layering 20 colors.

Let me know if you want to get into the technical DSP side of how they might simulate unison with low CPU — it's a fascinating area.

Is there any reason to believe that Hive actually does this, as in statements by Urs, or is this just speculation?

[kraster]

So it's kind of just wave-shaping like a Sawtooth wave to sound like it's being detuned is that right


ChatGPT said:
Yes — that's exactly the idea, and you're honing in on the core of what makes these synths efficient.

Here's the basic concept:

Instead of this:
You literally detune 8–16 separate sawtooth oscillators, each with their own phase and pitch offsets. This sounds great — thick, wide, and rich — but it's CPU-intensive, especially with polyphony.

They do this:
Plugins like Dune, Hive, and others often simulate that detuned sound using wave-shaping and clever unison algorithms. For example:

They might generate one sawtooth and then apply phase modulation, stereo spread, and detune simulation internally.

Some use interpolated waveform blending, where they morph the waveform shape over time or across a virtual unison spread.

Others use mathematical tricks to fake the sound of multiple oscillators by summing harmonics with controlled interference patterns.

So yes — it's as if they're just warping or shaping a sawtooth wave to behave like it's being detuned without actually playing multiple independent oscillators.

Bonus Analogy:
It's like a painter who makes it look like a complex texture with just a few brush strokes instead of layering 20 colors.

Let me know if you want to get into the technical DSP side of how they might simulate unison with low CPU — it's a fascinating area.

Is there any reason to believe that Hive actually does this, as in statements by Urs, or is this just speculation?

I think it's just speculation or rather discussion as to how Unison can be achieved using different techniques.

On the old Nord Modular in order to save CPU cycles they had two types of oscillators, a master and slave. The Master would receive all the control information and the slaves inherited that but you could detune them slightly. The Master Oscillator was considerably more CPU intensive than the slave.

One of the most used techniques in DSP is to use a shared waveform across all unison voices using a waveform lookup table and the applying different phases, tuning and panorama to that as copies. This allows you to use things like SIMD which will save CPU.

This gets messy when you have modulation and particularly things like warp modes.

In Serum 2 warp modes can have separate unison values to modify them. That would not be compatible with the kinds of unison that use the kinds of techniques above. You kind of have to render multiple discrete copies of each unison waveform copy

Dune 3, for example, doesn't really have any warp modes for the wavetables. Serum 2 has warp modes for every oscillator type.

[tumface]

tumface wrote:
There are some presets that are unusable, but I wasn't going to use those presets, anyway.

I primarily use Serum 2 on my Mac, where it can play those really heavy factory presets without any problems. I still don't use them, because I don't really use any of the factory presets. (I do use the factory wavetables and samples, though.) I don't usually have a need to desire to use presets like that. So, yeah.

[tumface]

Is there any reason to believe that Hive actually does this, as in statements by Urs, or is this just speculation?

No. It's made-up word salad from an AI. You could have put in a prompt for an explanation of how the wavetable synth Greebleglorp works, and it would have printed out something similar.

I really wish people wouldn't post the output of AIs into forum threads. It feels like pollution. If someone wants to read the output of an LLM, they can put in a prompt and read the output themselves. I think forums are for people.

[nanostream]

Is there any reason to believe that Hive actually does this, as in statements by Urs, or is this just speculation?

No. It's made-up word salad from an AI. You could have put in a prompt for an explanation of how the wavetable synth Greebleglorp works, and it would have printed out something similar.

I really wish people wouldn't post the output of AIs into forum threads. It feels like pollution. If someone wants to read the output of an LLM, they can put in a prompt and read the output themselves. I think forums are for people.

I appreciate that OP at least told us it’s an AI generated answer, which tells us we can stop reading because it’s just slop. It might be true, it might not be, we don’t know. But we do know for certain that’s it’s not a trustworthy source and a waste of time.

IMHO it doesn’t feel like pollution, it is pollution, and it seems to be getting worse. I think the best way to deal with it for now is to ignore posts like this and subtly ostracize people who think it’s a good idea to post meaningless word salad.

[teilo]

The reason for the efficiency of Dune and Hive (and probably Phase Plant) is that they aren't actually playing multiple oscillators, but are using wave shaping techniques to achieve the same result, while only needing to play a single or very few waveforms. But that same technique cannot easily be used in other oscillator types, such as sample and granular.

That's clearly not true, because Dune 3 and Hive 2 have the same level of efficiency when set to wavetable oscillator type. I'm not even sure how waveshaping would make unison oscillators work with any oscillator type, to be honest.

No so "clearly" as you presume. All audio transformation, regardless of kind, results in a new waveform. There is nothing special about unison in this regard. In the end, you have a new waveform that is the composite of all the others. I'm using "wave shaping" in a generic sense to refer to any method of altering a waveform to achieve an effect. You can wave-shape an oscillator to simulate a discrete filter circuit, for example.

Regardless, I've known this about Dune for at least 10 years. The unison is integrated into the oscillator itself. It's not summing individual waveforms.

And for the record, I have no opinion on the ChatGPT answer.

[tumface]

I'm using "wave shaping" in a generic sense to refer to any method of altering a waveform to achieve an effect.

I've never heard of anyone use 'wave shaping' to mean 'do anything with an audio waveform.' Well, yeah, by definition, Dune 3 is doing something with audio to make a unison effect. Because it's audio. I don't think a statement like that is too meaningful.

[teilo]

I'm using "wave shaping" in a generic sense to refer to any method of altering a waveform to achieve an effect.

I've never heard of anyone use 'wave shaping' to mean 'do anything with an audio waveform.' Well, yeah, by definition, Dune 3 is doing something with audio to make a unison effect. Because it's audio. I don't think a statement like that is too meaningful.

Wave shaping is any mathematical process in which a wave is processed into a distinctly different shape. It is achieved by applying non-linear transfer function to the original waveform. That function can be simple, or very complex. Folding, wrapping, and clipping are all examples of wave shaping. It's not just mapping one wave onto another.