Windows 10: audio dropouts on multi-core CPU setups

See on the Cubase forum.

FWIW, I have been through all of these mods and troubleshooting in search of better performance, less drop-outs etc on Windows 10. This is also in light of the fact that I’ve worked with mac os systems for many years, but now on a windows workstation because of the horsepower that is not available on Apple systems. The primary grunt issue here being 4k video production. In any case, perhaps the following experiences might be of use.

This dual Xeon box is pretty good: when its good, its extraordinary, when its bad its usually Windows or rather my lack of career length experience with Windows; getting better tho’. Cubase and Nuendo I suspect are also a little hamstrung by less-than-optimal multithreading code and the overheads seem to be greater than some other DAWs. In particular: Studio One 3 and Pro Tools 12.8 HD both of which perform far better in terms of load, throughput, latency and the rest. Still, I do prefer the overall experience of Cubendo.

So, in terms of improving performance for Cubeno, the following did not work:

  • BOIS settings, C-states off, Speed step, Hyperthreading etc made no difference (rather, only made the machine more noisy, hotter & consume more power.

  • The number of cores utilised for Win10 etc & the tricks suggested (as per this thread) made no difference.

  • audioengineproperties, ‘Set Affinity’ etc etc made no difference.

  • Otherwise, all of the other usual Windows tweaks for DAWs have been applied.

All of the above tweaks have been managed and tested periodically in the recent past.

Possibly, this workstation class PC has enough grunt and overhead so that most of the above does not apply [?] Perhaps the suggestions do make improvements in terms of squeezing the most out of less capable single core /gaming-type builds [?] … This machine is also maintained well and there are really no problems other than the odd Windows 10 ‘huh?’ but which tends to go away on the next boot or so. [Now running Creators Fall update, a few oddities there with the task bar sometimes, no big deal].

However, the following did make a difference in terms of performance increase and stability:

  • Cubendo settings: Steinberg Power plan on, Direct Monitoring on (w/ RME UFX+ buffer 128 & ‘Enable MMCSS for ASIO’ on), ASIO Guard standard settings, ie, nothing especially new here, but the Power Plan did help just a little.

3rd party apps, mildly improved performance, much greater stability:

So in my experience on this particular PC, the Bitsum tools made a greater difference than all the other stuff combined.

A final comment would be that Steinberg might consider the overall audio engine design in terms of better performance and multithreading in particular (likely coders working on exactly this stuff right now).

  • The UI as well. Presently there are a lot of settings spread about in various locations: some under ‘VST Audio System’, other under the audio interface settings, others in various places in the preferences. Perhaps these performance settings might be located all in the one place?

  • The CPU meter is very simple and ‘underdone’ by way of comparison with Pro Tools HD or Studio One 3. Perhaps this might be made far more detailed and accurate in terms of multi-core information, load, plugin impact, Input monitoring, MIDI thru to VIs etc. I found that tweaking and optimising overall performance as per above was infinitely more informative using these other apps. For example, by opening the Pro Tools Multiprocessor window it was easy to check various buffer setting against input record on/off load. Seems very accurate. Ditto Studio One & in terms of the dual Z latency arrangements for VIs and audio tracks.

I hope this may be of use.

I found the Bitsum tools to be very thoughtful and useful in relation to this particular topic. They have the usual free and pro versions; I ended up upgrading to the relatively inexpensive paid versions. Windows seems to be very happy with these & they also auto-manage profiles for other DAWs & NLEs. One nice feature is when not ‘pumping’ for grunt, they auto fall back to a Balanced, low power plan when idle.

Is it really that bad on the windows side?
I had planned to buy a new computer this fall and after 10 years of mac I was ready to change platforms because Apple had failed to build a computer that is the right one to replace my cheese graters.
Well Nu8 wasn’t what I expected and Apple finally announced the next gen Mac Pro, so I decided to wait but the Windows thing is not completely ruled out. But a post like this one really makes me think…

Unfortunately I think there is a bit of confusion in that thread in the Cubase section.

As far as I understand it Windows will give an app a maximum amount of logical cores for ‘multimedia’, which is what the DAW is (to the OS). With HyperThreading on you get two logical cores for each physical. So according to the Steinberg support page when you have more than 14 logical cores you might see performance degradation.

So first of all that means 8 cores or higher in order to maybe experience a problem. Many people still run quad- and hex-core CPUs. Secondly, if what Steinberg says is true you could get 8 cores or more and limit the usage by Nuendo to 7 of those cores. Now, intuitively that may seem like a bad idea since we’re now “wasting” the other cores, but they probably won’t sit idle anyway and instead actually do work we care about. So it might not be as bad as one thinks. Thirdly, some people just buy a computer or CPU and don’t overclock it. If the CPU ends up using turbo because it’s the fastest way of processing DAW data then typically that won’t be on all cores anyway. So again one might wonder if the performance hit is really that big. One would have to overclock all cores to have a relative disadvantage I would think. And lastly I have to wonder about buffer sizes and if they make a difference on individual systems.

Anyway, I’m not sure just how big this issue really is.

Interesting. I have a heavy Xeon setup with dropouts in Cubase all the time and Bitsum changed absolutely nothing for me.