Zebra 3 and its CPU Burner

Official support for: u-he.com
Post Reply New Topic
RELATED
PRODUCTS

Post

In case you missed it, and because we forgot to mention it in the preliminary user guide:





Note that we'll probably move the Burner from the FX Grid to the Module Performance Panel. We'll see...

Also, as it came up a few times:

- the CPU Burner is only used *temporarily* to get reliable readings
- we can't just add a button to follow all suggestions because they can change the sound
- there'll be more suggestions, e.g. "use the Vanilla filter instead of Ladder" (different sound!)

Post

Truly awesome, Urs.

Post

As i understand cpu burner is relevant to new Intel and Apple's M silicon chips? What about ryzen series?

Post

Lbdunequest wrote: Fri Dec 19, 2025 6:02 am As i understand cpu burner is relevant to new Intel and Apple's M silicon chips? What about ryzen series?
I had to ask a computer about this:

"No, current Ryzen CPUs don't use the asymmetric efficiency/performance core architecture that Intel adopted with their 12th gen (Alder Lake) and newer processors.

AMD's Ryzen CPUs use a homogeneous core design - all cores are the same type. Even their latest Ryzen 9000 series (Zen 5) processors have uniform cores, though they do implement different power states and boost behaviors across cores.

However, AMD does have some asymmetry in their designs:

- **Zen 4 and Zen 5 desktop chips** have cores grouped into CCDs (Core Complex Dies), and the OS can show preference for certain CCDs based on which are of higher quality bins
- **Ryzen mobile chips** have used different approaches - some combine Zen cores with older architecture cores (like Zen 4 + Zen 4c in their "Phoenix" APUs), where the "c" variant is denser and lower power but still the same architecture

For your DSP work on Burn module CPU optimization, this means Ryzen systems won't have the same scheduling complexity as Intel's hybrid architecture. All Ryzen cores can run at full performance, though you might see some variation in boost clocks between CCDs on multi-CCD parts like the Ryzen 9 chips.

The Intel hybrid architecture is really the outlier here requiring special thread scheduling considerations for real-time audio processing."

If it hasn't hallucinated, I'd say the CPU Burner is not relevant for Ryzen CPUs at this moment.

Post

This is interesting, I haven't used Logic for this, but I made a patch with Z3, one osc, filter, adsr, reverb. Brought my M2 Studio Ultra to it's knees in Digital Performer 11, which usually has good load balancing, but does use efficiency cores. It's possible I can force better performance using the CPU Burn to move it to a single performance core? Logic from every test I've done does not use efficiency cores for anything with plugins, so I'll test that in a couple minutes here and report back.

To be honest though, this is the first plugin with a relatively simple patch that crushes the M2 just by playing multiple notes. Hopefully there is some optimization happening in the future or anyone with a three year old machine will run into issues.

Post

machinesworking wrote: Fri Dec 19, 2025 6:30 am To be honest though, this is the first plugin with a relatively simple patch that crushes the M2 just by playing multiple notes. Hopefully there is some optimization happening in the future or anyone with a three year old machine will run into issues.
Depends. The Yellow filter for instance easily uses 10% per instance per voice on a modern computer. It's an extreme filter based on a very complex model of an analogue filter. It has already been optimised ad nauseam, one developer put a lot of effort into this, and nothing we can still do would keep its sound intact.

However, it obviously sounds marvellous ask it oozes character. But for a polyphonic patch, one should compromise and use the SVF instead, maybe add a distortion after.

(this would be a classical application for the Module Performance Meter, to find compromises that use less CPU but preserve the overall character of the sound)

Post

Urs wrote: Fri Dec 19, 2025 6:40 am Depends. The Yellow filter for instance easily uses 10% per instance per voice on a modern computer. It's an extreme filter based on a very complex model of an analogue filter. It has already been optimised ad nauseam, one developer put a lot of effort into this, and nothing we can still do would keep its sound intact.

However, it obviously sounds marvellous ask it oozes character. But for a polyphonic patch, one should compromise and use the SVF instead, maybe add a distortion after.

(this would be a classical application for the Module Performance Meter, to find compromises that use less CPU but preserve the overall character of the sound)
Yeah that's exactly it, I could not for the life of me find the patch I made but testing now I'm certain I gravitated to that filter considering the weird feedback it can get! I suspect I closed the test project and hit don't save. The Yellow Filter is a pig! I'll keep that in mind! :lol:

I was not able to replicate killing the CPU again, unless I used the Burn at 100. FYI, in my tests Reaper and DP11 were the highest track count DAWs, Logic Bitwig and Live got less instances of Diva in that order. But, it seems that DP11 addresses cores and Mac OS differently than Reaper, despite identical track counts? as I could not force Zebra to jump to performance cores with the Burn, but again no overloads so....

Post

koalaboy wrote: Thu Dec 18, 2025 3:16 pm Truly awesome, Urs.
Indeed, Indeed and Indeed. 8)

Post

Urs wrote: Thu Dec 18, 2025 3:06 pm In case you missed it, and because we forgot to mention it in the preliminary user guide:

This is great! I understand this P core/E core issue fairly well, being a steinberg user for a long time and running into those exact problems for years after the new CPU/chipsets were released. Its very cool how you built the CPU burner to 'kick' the threads over to the P core!! I might need to use this in cubase15 sometimes.

Post

I really really really think one should only use this for measuring purpose.

The final version will certainly switch the CPU Burner off when not measuring.

However, it can probably be used to demonstrate to other software developers that there might be a problem that can be resolved.

Post

Urs wrote: Mon Jan 05, 2026 10:47 am

However, it can probably be used to demonstrate to other software developers that there might be a problem that can be resolved.
Understood 100%. Thanks. I only use this for momentary testing. It becomes a tool in my toolbox. ; )

Post Reply

Return to “u-he”