Audio dropouts in Cubase 12.0.20

I am literally running an empty project and playing the Classic Sweeper patch of the Roland Cloud JD-800 in realtime and I am experiencing audio dropouts when I repeatedly press a 3 note chord.

Any ideas what could be causing these dropouts on such a basic project?

Core system
Windows 10 Pro 21H2 build 19044.1682
M-Audio M-Track Eight USB audio interface
M-Audio Oxygen Pro 49 (latest firmware)
Intel(R) Core™ i7-9700K CPU @ 3.60GHz 3.60 GHz
ASUS Prime Z-390P motherboard (latest BIOS)
32GB Crucial RAM
2TB WD Blue SSD drive (C drive)

Activating Audio Power Scheme seems to have helped a lot

That’s perfectly normal and is not related to Cubase.
When you play the same chord repeatedly, the number of voices that are played simultaneously rapidly increases and overloads the realtime processing.
To prevent this, most VST instruments have the ability to set a maximum polyphony limit.
I don’t know about Roland but that’s usually the setting to you should find and tweak first when your CPU isn’t fast enough. (Or the synth is too demanding, you get the idea)

Activating the Steinberg Audio Power Scheme is basically the same thing as the Windows High Performance power plan, apart from a few minor things.
That will simply put your CPU to its maximum clock independently of the load.

But, it all depends on your actual CPU settings. The best is to leave the Power Plan on Balanced in Windows, and do the right setting in the BIOS.
The good thing is that we have the same CPU and Chipset.
I don’t know if the settings are labelled the same nor where they are located on an Asus motherboard (because I have an MSI), but here’s my settings :
What’s in bold was disabled or set to a different value by default.

  • Clock/Ratio settings = Auto
  • Intel Turbo Boost = Enabled
  • Enhanced Turbo = Enabled
  • EIST = Enabled
  • Intel Speed Shift = Enabled
  • Intel C-State = Disabled
  • CPU Core Voltage Mode = Auto
  • CPU Core Voltage = If you leave it on Auto it is too high. It is better to set a value manually. For me 1.250 V is working good. (for information, Auto is 1.38 V under load)
1 Like

What makes you say that? I have never seen anyone recommend a Balanced power plan in any version of Windows. I would highly recommend against it! Why would you want the OS to arbitrarily put components in a low power state?

Because he is disabling the c-states of the CPU in the BIOS, which basically prevent the CPU from going into power saving mode.

The disadvantage is that your CPU is running full power all the time. If you use the PC exclusively for DAW work, that might be fine, but if it is a multi purpose system, not unusual with hobbyists like me, it is not what I would want (I have a dedicated power plan which I activate via a batch file for DAW work).

Windows power plan governs more that just the CPU power. I still highly recommend against it.

I know that the High Performance power plan does not only affect the CPU frequency, but also SSD/NVMe power management and many other things that can increase the performance.

Would you agree to make a test ?

With the defaulted power plans :
Set the power plan on Balanced and set the Minimum Processor State to 100% so the CPU clock is always at its maximum.
Open whatever project in Cubase or run any benchmark program.
Now set the power plan on High Performance, and compare Cubase performance or benchmark scores.
Do you see any noticeable difference between the two power plans ?
If you do, then I’m really impressed.

If we follow the same logic, the Ultimate Performance power plan is supposed to give even better performances than the High Performance one. But no one would actually see the difference in a real world setup.

No, disabling C-States doesn’t make the CPU run at full power all the time.
It only disables the CPU power saving features.

Basically, C-States are only really useful when the computer goes into sleep, when the screen turns off, or when you let the computer idle and there is almost no load.

Of course C-States also work when you actually use the computer under certain types of load, but then this is more of a passive thing to help save power even more.
The C-States won’t be as deep as in the situations above.
A core can enter and exit soft C-States many times per second, but doing so will cause the core to be unavailable for the time it takes to “wake up” when it gets told to work. This feature indeed steals some CPU cycles and can lead to lower performance.

Since Cubase usually has a relatively low CPU usage (I mean even with a 50 tracks project, the load is very far from a constant 100% usage like 3D renders), disabling C-States can for sure help, but the improvement will be very slight, if not absent in most cases.

Disabling C-States isn’t a magical performance solver.

The ability for the CPU clock to change is not related to the C-States, but to EIST (Enhanced Intel SpeedStep Technology) and the more recent Speed Shift.
For the CPU frequency to be able to move, either of EIST or Speed Shift may be enabled.

For the best reactiveness concerning the speed at which the CPU clock can change, I highly recommend to enable both EIST and Speed Shift.
Speed Shift is the most important setting you can actually enable. The CPU Clock will react extremely fast to any load, and will not cause any performance issue.

Speed Shift is extremely well implemented into Cubase, if not the best implementation in any software I’ve used, so there is absolutely no reason not to allow dynamic clock with Cubase.

It is possible to get the CPU to react to the load, even with the High Performance power plan.

The adaptive clock in relation to the load is governed by a hidden Power Plan setting called “Processor energy performance preference policy”, and the usual Min/Max Processor State settings, which simply set the clock limits.

This hidden setting is the Speed Shift fine-tune, and acts like a soft knee threshold where the CPU clock can change depending on the load.
When High Performance power plan is selected, this setting is set to 0 by default, and that is why the CPU will work at max speed, even when there is no load.
To make it visible in Power Plan settings, you must go into the registry, search for this entry and change the Attributes value to 2 :


Now it is visible, you can go into your High Performance power plan settings, and change the value to somewhere between 30 and 40.
It depends on the number of background tasks you have during idle, so experiment accordingly, and use a monitoring graph when you do so, for example Afterburner/RTSS.

However, don’t forget to set the Minimum Processor State to 0%, don’t keep it at 100% or the clock won’t be able to move.

With only the default background programs running, and without touching your mouse (idle), you should aim for a flat line on minimum CPU clock.
It may do some spikes but that’s not a problem, you even should allow it to make some spikes from time to time during idle, for example when Windows fetches updates and does its things, and most importantly when you move the cursor over the desktop.
The latter would confirm that the system will be as responsive as possible to any load, and at the same time reducing power consumption when there isn’t.

1 Like

Thanks for the detailed information! I am definitely not a hardware guy, and it shows :wink: i did actually write a specific cpu stress test program once for testing our servers, but those had pre-defined workload profiles which set a shitload of values I have no idea about… the results were interesting. Your regular cpu stress test apps are rather useless, because they usually just max out all cpu which is not real world, so my program simulated some intermittent pauses between calculations. Then there were measurable differences between different CPU BIOS settings.
Sorry, OT…

1 Like

Yes, I agree that when the workload is highly fragmented, the different settings can exhibit noticeable differences.