Zebra 3 and Zebra Legacy

[gentleclockdivider]

[...]

Despite all this, I would love to see an "FM from Osc2" or better "FM from Sidechain" as an Osc-FX. "FM from Sidechain" would be pretty neat --> this would open up a whole can of whoop ass.

BTW, I am asking for the Yamaha kind of FM --> I am talking about PhaseModulation (like the DX Osc-FX).
In MSoundFactory you have the possibilty of both with such kind of "complex" routings and I prefer PM over FM.

Zebra fm modules are phase mod

[exmatproton]

Despite all this, I would love to see an "FM from Osc2" or better "FM from Sidechain" as an Osc-FX. "FM from Sidechain" would be pretty neat --> this would open up a whole can of whoop ass.

+1

[gunhed]

Quick question!

I know if I purchase Zebra Legacy there's no upgrade path towards Zebra 3. But I noticed a handful of retailers still have the original Zebra 2 listed for sale (at the original price). Would purchasing this put me in the $30 upgrade bracket?

[zvenx]

Until you get the official word from a u-he employee, I believe the answer is yes.
rsp

[Urs]

Quick question!

I know if I purchase Zebra Legacy there's no upgrade path towards Zebra 3. But I noticed a handful of retailers still have the original Zebra 2 listed for sale (at the original price). Would purchasing this put me in the $30 upgrade bracket?

Depends on what they pay us, I guess. If they still ask the old price (which of course they can) but then only send us our share from the new price, I don't really know how we would be supposed to compute that. Since we only ever get the new price now - distributors and stuff - and since we absolutely get no information about what the retailer asks of you, I would not go for such "offer".

Most likely it is a shop which has not been in touch with our distributors who had a several months notice before the fact.

[lokio]

Hello,

If possible, I'm interested to know what sort of hardware (CPU instructions etc) Z3 will require.

Lokio

[Urs]

Hello,

If possible, I'm interested to know what sort of hardware (CPU instructions etc) Z3 will require.

Lokio

Atm we develop for SSE 4.2 or Neon as minimum. It would be tempting to go for AVX2 as minimum for Intel compatible machines, but then it would also require two separate code paths. Hence it isn't likely that we switch from 128 bit SIMD to 256 bit SIMD as long as Apple's CPUs do not support SVE on Arm architectures (there is a trick to keep a homogenous code base, but I'm not sure if it's with the effort.)

Hence: SSE 4.2 and Neon in the foreseeable future.

For Zebra3 I have experimented with a way to offer alternative support for AVX, AVX2 and AVX512. AVX doesn't do much, it has too many penalties. AVX2 and AVX512 however scale directly for some algorithms, such that certain code will run 2x or 4x faster depending on the feature set of the CPU. The same binary will however also execute fine with SSE 4.2 only.

[Fannon]

Just curious - how much end user performance is there to gain realistically by cutting support for those old instruction sets? Unfortunately, I don't know much about this. Wouldn't the compiler be able to create code that makes use of newer instruction sets -if available- and if not, fall back to something slower?

[Urs]

One can't just mix code for different instruction sets. The switching of back and forth would cost a lot of time.

One common thing to do, one compiles the whole code several times, once for each instruction set, and the offers separate installers or the installer figures out which one to install. This is a quick way to benefit if the developer leaves it all up to the compiler, but the benefits are maybe - I'm speculating - in the realm of 10%-20% more efficiency for what we do.

Another method to support different instruction sets is to compile individual parts of the code for these different instruction sets and then selects those parts that apply while the software is running. This is suited for algorithms which are hand coded with assembly-like optimisations. In my experience this gives individual algorithms a 100% - 400% better efficiency.

While compilers are supposed to be able to do "auto vectorisation", I have not been able to make this work for anything but the most simple algorithms. Even when copying samples from one buffer to another while interleaving or deinterleaving, my hand written assembly-like code was a lot faster that what a compiler came upon with thus far.

[Fannon]

Thanks for the detailed answer, @Urs!

[TENETfanboy]

Any update on the Zebralette 3?

[Urs]

Any update on the Zebralette 3?

Not much has happened from my side in recent months - I did a lot of bug fixing instead, and I worked on Filterscape improvements that will also leverage into Zebra3 and Uhbik, but not Zebralette3 (some of these improvements have used the new waveform editor though, so I did actually work with it - and a little bit on it).

However, other devs have created and prepared some new tech for it, which took nearly a year to develop. It converts arbitrary samples to wavetables and it can choose the most "different" waveforms to make available in the curve based editor. This is about to be merged into our uhm scripting engine which then is available to Zebralette3's spline based waveform editor.

As for roadmaps, I hope to be ultimately finished with Filterscape well before the NAMM show and I hope to have a nearly feature complete version, possibly an alpha, of Zebralette3 for Superbooth.

[Fannon]

So we'll get a free Zebralette3 first and then a Zebra 3 (beta?)?

[Urs]

So we'll get a free Zebralette3 first and then a Zebra 3 (beta?)?

That was always the plan. Zebralette is a subset of Zebra, and a useful tool to learn some core concepts of Zebra, i.e. waveform/MSEG editing, oscillator effects...

[TENETfanboy]

Any update on the Zebralette 3?

Not much has happened from my side in recent months - I did a lot of bug fixing instead, and I worked on Filterscape improvements that will also leverage into Zebra3 and Uhbik, but not Zebralette3 (some of these improvements have used the new waveform editor though, so I did actually work with it - and a little bit on it).

However, other devs have created and prepared some new tech for it, which took nearly a year to develop. It converts arbitrary samples to wavetables and it can choose the most "different" waveforms to make available in the curve based editor. This is about to be merged into our uhm scripting engine which then is available to Zebralette3's spline based waveform editor.

As for roadmaps, I hope to be ultimately finished with Filterscape well before the NAMM show and I hope to have a nearly feature complete version, possibly an alpha, of Zebralette3 for Superbooth.

Awesome thanks for the update!! Can’t wait to see it!! Also very excited to hear these new filters