Official Serum 2 thread!

[dionenoid]

Here's the dev's answer on all of these "cpu issues" :

There's a lot of threads on the subject, maybe a comprehensive post is needed.

Serum 1 shipped with a fair number of CPU complaints also, and over time they became basically none. Of course I hope that to be the case here. But I do think part of that reason was computers typically getting faster, people in 2014 were using computers from 2007 for instance, and now not so much.

Serum 2 has been optimized extensively however it is also very large and conditional system, so there are further gains to be found I'm sure. However I think it's safe to say that low hanging fruit / major gains have been already realized. I'm actually very proud of the performance in places, it isn't so obvious if you don't see what "normal/typical" would look like.

You may be running a computer with lower specs, or there could be gains to be had possibly with audio driver or settings (e.g. raising the process buffer range in Logic might help). Running in Good quality mode is still quite good. Ultra is overkill for most any need.

You should see Serum 2 performance on par with Serum 1 when doing Serum 1 things (e.g. Serum 1 presets) and it is "doing more" as the quality of Serum 2 is higher inherently. So it is actually higher quality and higher performance already. I hate to have to say it, but most synths cut a lot of corners we are not cutting, because having the best quality synth possible is of primary interest, and given that the CPU load is dependent on what you ask of it, it sort of makes it not a consumer level product / a little understanding of voice count and the costs associated is of benefit.

Spectral and Granular can both take a lot of CPU which is not a surprise knowing what they are doing. if you don't keep in mind what you're asking of the computer. For instance unison on both of these should probably be avoided in polyphonic situations as it makes CPU 2x, 3x, 4x etc and the sound benefit is unlikely to be there in polyphonic situations anyway.

The VA (new) filters also take a lot of CPU as they are mostly component modeled (DJ Mixer and PZ Filter, the latter of which requires a lot of CPU also inherently). Sometimes the tradeoff of using a component modeled filter is not worth it / inaudible in some contexts.

I do think it's worth understanding the cost associated with various stuff, watching Poly in the lower-right and specifically the numerator of the fraction lets you know how many voices you are asking of your CPU. Unison counts / grain counts, long Env 1 release, and a lot of notes/poly can push this number high.

We are putting further time and focus on this - Things will only get better over time, we are making both fixes and improvements, the list of changes coming in the next update is pretty long, and as the attention on bugs is getting behind us we can really dive down in to specific cases and any hotspots that we're able to address or improve on. There are a couple things which can improve performance in the next update already, however the base CPU is still what it is.

And :

If there is a filter on the main panel, but it isn't really doing something per-note (e.g. keytrack, or envelope) or routed discrete to an oscillator, you might be better off moving it to a Filter FX on the main bus for instance. This way there is One (Stereo) filter instead of (# voices *2 stereo) filters running. Especially with the new filters (which are mostly component modeled) this can save a whole lot of CPU (perhaps the majority) without changing the sound.

(Both quotes from Steve Duda. Posted at the official Xfer forums)

[SirKen]

Here's the dev's answer on all of these "cpu issues" :

There's a lot of threads on the subject, maybe a comprehensive post is needed.

Serum 1 shipped with a fair number of CPU complaints also, and over time they became basically none. Of course I hope that to be the case here. But I do think part of that reason was computers typically getting faster, people in 2014 were using computers from 2007 for instance, and now not so much.

Serum 2 has been optimized extensively however it is also very large and conditional system, so there are further gains to be found I'm sure. However I think it's safe to say that low hanging fruit / major gains have been already realized. I'm actually very proud of the performance in places, it isn't so obvious if you don't see what "normal/typical" would look like.

You may be running a computer with lower specs, or there could be gains to be had possibly with audio driver or settings (e.g. raising the process buffer range in Logic might help). Running in Good quality mode is still quite good. Ultra is overkill for most any need.

You should see Serum 2 performance on par with Serum 1 when doing Serum 1 things (e.g. Serum 1 presets) and it is "doing more" as the quality of Serum 2 is higher inherently. So it is actually higher quality and higher performance already. I hate to have to say it, but most synths cut a lot of corners we are not cutting, because having the best quality synth possible is of primary interest, and given that the CPU load is dependent on what you ask of it, it sort of makes it not a consumer level product / a little understanding of voice count and the costs associated is of benefit.

Spectral and Granular can both take a lot of CPU which is not a surprise knowing what they are doing. if you don't keep in mind what you're asking of the computer. For instance unison on both of these should probably be avoided in polyphonic situations as it makes CPU 2x, 3x, 4x etc and the sound benefit is unlikely to be there in polyphonic situations anyway.

The VA (new) filters also take a lot of CPU as they are mostly component modeled (DJ Mixer and PZ Filter, the latter of which requires a lot of CPU also inherently). Sometimes the tradeoff of using a component modeled filter is not worth it / inaudible in some contexts.

I do think it's worth understanding the cost associated with various stuff, watching Poly in the lower-right and specifically the numerator of the fraction lets you know how many voices you are asking of your CPU. Unison counts / grain counts, long Env 1 release, and a lot of notes/poly can push this number high.

We are putting further time and focus on this - Things will only get better over time, we are making both fixes and improvements, the list of changes coming in the next update is pretty long, and as the attention on bugs is getting behind us we can really dive down in to specific cases and any hotspots that we're able to address or improve on. There are a couple things which can improve performance in the next update already, however the base CPU is still what it is.

And :

If there is a filter on the main panel, but it isn't really doing something per-note (e.g. keytrack, or envelope) or routed discrete to an oscillator, you might be better off moving it to a Filter FX on the main bus for instance. This way there is One (Stereo) filter instead of (# voices *2 stereo) filters running. Especially with the new filters (which are mostly component modeled) this can save a whole lot of CPU (perhaps the majority) without changing the sound.

(Both quotes from Steve Duda. Posted at the official Xfer forums)

Thank you for sharing! I cannot agree with Steve there as there seems to be a performance issue with AMD CPUs overall. In any case, I am glad I held back from purchasing before hearing any commitment to address the concerns. And with that satisfying closure, I will leave the Serum 2 discussion to people who are enjoying it on their M series chips.

[Uppity]

I'm mostly interested in the "clips" feature in Serum 2, offering polyphonic sequencing, like having a daw within a synth. I noticed there's a button for "gate" in "clips", but I'm not clear if it works the same way as in most arpeggiators, in which it can sharply snip off the end of notes to achieve a staccato effect. There's no knob to adjust it. Is this "gate" just for snipping off the end of a phrase rather than individual notes? Has anyone out there played with this and can comment?

[JO512]

EDIT: deleted. I need to test this some more to be sure.

[jtsterays]

What higher quality is he talking about? A pure saw chord with no filter in S2 is 2x the CPU compared to other synths while sounding the same.

[Gam456]

Thank you for sharing! I cannot agree with Steve there as there seems to be a performance issue with AMD CPUs overall. In any case, I am glad I held back from purchasing before hearing any commitment to address the concerns. And with that satisfying closure, I will leave the Serum 2 discussion to people who are enjoying it on their M series chips.

Sure I can run 128 Diva instances.
But I'm also able to run 16 Serum 2 instances with heavy preset @48 khz - 64 buffer on my Ryzen 7950X.
It's all about optimisations.

[pierb]

What higher quality is he talking about? A pure saw chord with no filter in S2 is 2x the CPU compared to other synths while sounding the same.

Probably oversampling. You can adjust the global quality in the global settings. By default it's set to high which iirc is 2x oversampling.

[pierb]

I pretty new to Serum and a bit confused as to how the wavetable engine works.

The wavetable editor seems very much like an additive oscillator editor similar to Zebra although I'm 99.99% certain the wavetable engine is not an additive engine (probably what the spectral engine does).

So what is Serum using at runtime? Does it convert a single cycle from the editor o imported PCM audio into like a math formula/equation? Or is it a different format more like vector graphics bezier curves and points?

I checked the manual and couldn't find much information about this. Hoping someone here is an expert and can clarify how it works internally!

[Funky40]

my thoughts:
first of all, Steve should tell his instructions to the preset makers ! Seems there went not enough love into their work in that respect.
2nd, add a CPU% load reference -of some sort- to the preset names.
I think this could help the userbase alots for now.


i can run one core on my M4studio-max with one S2 sound into its knees easily, if i don´t watch how i hit the keys.
Since my Host runs on one core only, well....the host works good as a CPU load reference, haha.
(refering to the sounds i used to play with, since i liked these)



I support the idea to create just the best possible vs. SW-Synth creation. I like that idea.
But give instructions to the users. Thats a good idea !
i think right now, that´s probably the more important "work" to do, than to work on code optimisations.

Make a clear CPU optimisation manual please vs. preset creation !.....no ?
Edit: rather than a random post in the middle of a soon even longer thread.
But thanks for that post !

[Alchemedia]

I'm mostly interested in the "clips" feature in Serum 2, offering polyphonic sequencing, like having a daw within a synth. I noticed there's a button for "gate" in "clips", but I'm not clear if it works the same way as in most arpeggiators, in which it can sharply snip off the end of notes to achieve a staccato effect. There's no knob to adjust it. Is this "gate" just for snipping off the end of a phrase rather than individual notes? Has anyone out there played with this and can comment?

What do you see as the advantage over simply using another seq/arp or your DAW unless you are selling one finger presets?

[Korg Supporter]

I pretty new to Serum and a bit confused as to how the wavetable engine works.

The wavetable editor seems very much like an additive oscillator editor similar to Zebra although I'm 99.99% certain the wavetable engine is not an additive engine (probably what the spectral engine does).

So what is Serum using at runtime? Does it convert a single cycle from the editor o imported PCM audio into like a math formula/equation? Or is it a different format more like vector graphics bezier curves and points?

I checked the manual and couldn't find much information about this. Hoping someone here is an expert and can clarify how it works internally!

Serum wavetables have 2048 samples for the waveforms. So they are basically arrays of short samples. When you are using the editor, you are manipulating the harmonic conent in real-time. However, the waves are not generated by partials. Vital and Peak have wavetables generated via partials.

[kraster]

I pretty new to Serum and a bit confused as to how the wavetable engine works.

The wavetable editor seems very much like an additive oscillator editor similar to Zebra although I'm 99.99% certain the wavetable engine is not an additive engine (probably what the spectral engine does).

So what is Serum using at runtime? Does it convert a single cycle from the editor o imported PCM audio into like a math formula/equation? Or is it a different format more like vector graphics bezier curves and points?

I checked the manual and couldn't find much information about this. Hoping someone here is an expert and can clarify how it works internally!

It's a wavetable!

I think there's slight confusion with the term wavetable.

I think there's a common tendency to think of wavetable as a collection of different waveforms that are scanned through.

A wavetable in the purest form is an array of samples values that represent 1 cycle of a periodic waveform that is precomputed and stored in memory.

So in the editor you're essentially just defining the array of one waveform.

You are only ever editing one frame (sub table) in the editor.

Serum uses an FFT based system where the bars in the window represent the harmonic content of the waveform. It's essentially an additive style interface.

The lower bars represent the phase offset.

[kraster]

I'm mostly interested in the "clips" feature in Serum 2, offering polyphonic sequencing, like having a daw within a synth. I noticed there's a button for "gate" in "clips", but I'm not clear if it works the same way as in most arpeggiators, in which it can sharply snip off the end of notes to achieve a staccato effect. There's no knob to adjust it. Is this "gate" just for snipping off the end of a phrase rather than individual notes? Has anyone out there played with this and can comment?

The gate just allows the sequence to be passed to the oscillators.

It's no an audio gate though.

By default the sequence is cycling around and activating gate allows the note information to be passed to the synth.

You can also have gate reset so it plays from the beginning each time you press a key.

[swilow11]

I pretty new to Serum and a bit confused as to how the wavetable engine works.

The wavetable editor seems very much like an additive oscillator editor similar to Zebra although I'm 99.99% certain the wavetable engine is not an additive engine (probably what the spectral engine does).

So what is Serum using at runtime? Does it convert a single cycle from the editor o imported PCM audio into like a math formula/equation? Or is it a different format more like vector graphics bezier curves and points?

I checked the manual and couldn't find much information about this. Hoping someone here is an expert and can clarify how it works internally!

Serum wavetables have 2048 samples for the waveforms. So they are basically arrays of short samples. When you are using the editor, you are manipulating the harmonic conent in real-time. However, the waves are not generated by partials. Vital and Peak have wavetables generated via partials.

In the context of additive synths like Razor, partials refers to discrete sine waves. Vital doesn't use that sort of oscillator.

[Korg Supporter]

I pretty new to Serum and a bit confused as to how the wavetable engine works.

The wavetable editor seems very much like an additive oscillator editor similar to Zebra although I'm 99.99% certain the wavetable engine is not an additive engine (probably what the spectral engine does).

So what is Serum using at runtime? Does it convert a single cycle from the editor o imported PCM audio into like a math formula/equation? Or is it a different format more like vector graphics bezier curves and points?

I checked the manual and couldn't find much information about this. Hoping someone here is an expert and can clarify how it works internally!

Serum wavetables have 2048 samples for the waveforms. So they are basically arrays of short samples. When you are using the editor, you are manipulating the harmonic conent in real-time. However, the waves are not generated by partials. Vital and Peak have wavetables generated via partials.

In the context of additive synths like Razor, partials refers to discrete sine waves. Vital doesn't use that sort of oscillator.

Sorry if I remembered. I thought Vital used some kind of spectral wavetables.