Windows 10: audio dropouts on multi-core CPU setups

Iā€™'ve switched because Iā€™m a hardware GEEK & NERD and love being a beta tester for new setups and configurations :smiling_imp:

Iā€™ve been provided finally, after two weeks or so, by Polish Yamaha (acting as Steinberg distributor) with a registry fix that brought nothing but sour happiness and peace of mind, sort of anyway. My impressionsā€¦

Not everything runs great. Sometimes sound stutters, sometimes load peak goes crazy with no reason and sometimes itā€™s caused by plugins like Kontakt 5 (heavily armed instances) and for example Kinetic Metal or by Spectrasonics Omnisphere 2.Have no problem with my SSL Duende plugins, they run fine. Iā€™ve disabled some features of my CPU in BIOS (but HT is on, Windows 10 wo HT is pain in the aā€¦) and Iā€™m sure that I have a moderately clean Windows 10 Pro enviroment. Iā€™m currently using RME UCX but it never gave me any problems, DCP latencies are low. I miss Cubase 6 and the old times, because I believe that this DAW run smoother and better back then (Iā€™m using C since first VST versions in late 90s). Or maybe Iā€™m just old and grumpy :smiling_imp: :smiling_imp: :smiling_imp:

What Iā€™ve experienced leaving HT on creates VST performance spikes of 20% or more. Once HT is turned off average VST performance usage stayed the same but the realtime spikes disappeared.

From what Iā€™ve been reading, HT doesnā€™t benefit realtime processing like DAWā€™s, except rendering audio files. Running low buffer setting will be a lot more noticeable at high buffer settings

Iā€™m interested in what other people are experiencing with HT on vs HT off.

My setup
Asus z370-A motherboard, 8700k cpu, 32GB ram, NVMe 970 evo 500GB, NVMe 970 EVO 1TB, 2 x seagate 2TB 7200rpm hdd, GTX560, Windows 10 pro, Motu 828mk3 hybrid via pcie firewire and Cubase Pro 9.5.30. Running a 192 buffer at 44.1 with ASIO guard off.

I have the same CPU and disable C-States in the UEFI/BIOS and the Steinberg engine proporties seems to work and give less spikes and more stable audio processing. not that Iā€™ve had any dropouts at 256 ms buffer and 32bit 96khz less spikes on the meter. c-states disabled really gives a difference.
x299 7820x @ 4.6 Ghz around 1.2ish volts HT on. (other options to disable like EST and turbo boost donā€™t change a thing than overall power consumption on this platom so why bother.)

Thx for the tip. Will test that very soon.

Hello from Russia :slight_smile: So I have very similar problem on my AMD Ryzen 7 1700 CPU. I have a project in Cubase that uses 5 Spires, 5 Kontakts and 1 Omnisphere and with ASIO Guard OFF itā€™s not running well on 128 (crackles and dropouts), only with 512 ASIO buffer size works kinda fibe. While in FL Studio the same (carefully replicated by MIDI and presets) project works flawlessly on 128 buffer. Turning ON ASIO Guard makes a huge difference tho, itā€™s possible to work at 128 with no crackles and dropouts and CPU coreā€™s load is pretty even, but I expected much more better performance from Cubase 9.5 Pro though.

Where should I start? Turning off HT in BIOS? Or writing Fabil a PM? Does this problem apply to Ryzens as well as Intels?

Hy Guys,

May be already mentioned in the discussion :

Try to install Asio4All

It solved my sound problem with windows 10 and cubase pro.

Good luck :wink:

So, I used fix from Fabio. I think there is some kind of improvement: with ASIO Guard my averagely loaded project at 256 buffer size having less load on CPU (reduced to 19% from 30% somehow and the load on cores/threads is very even and smooth) and selecting the track for realtime record doesnā€™t make +50% CPU load but only ~15%. But performance without ASIO Guard on 256 ASIO buffer remained pretty poor and itā€™s still even worse than FL Studioā€™s :frowning:(
Though with ASIO Guard itā€™s decent enough to continue my work.

To be honest I donā€™t think itā€™s a correct solution. I have RME Babyface Pro and a lot of features are available through itā€™s driver, furthermore, some developers like RME or Lynx are making excellent and super-stable ASIO drivers and it would be an absolute waste to not use them.

I agree with you JafaroffMusic : itā€™s always better to use the dedicated driver, But it was the only working solution in my case.

Got it :slight_smile:

Are ASIO4ALLā€™s peformance at 128 buffer size and dedicated driverā€™s performance at 128 buffer size comparable? Iā€™m asking this because I still have some kind of doubts on this problem - is it software-related or hardware-related or both? In my case Cubase is able to load ton of SGA plugins which are used in DAWBench DSP test - my Ryzen 7 1700 usually begin to produce dropouts only after 115th plugin at 44.1 kHz and 64 ASIO bufferā€™s size leading to overall CPU load around 81%, the load percentage I never saw on Cubase 9 with other plugins that I use everyday - Kontakt, Spire and etc. Why my CPU is able to hold so much SGAā€™s and is not able to hold just a few Kontakt/Spire/DUNE/Twin2 instances? Why FL Studio is able to hold this (without ASIO Guard, AFAIK FL Studio doesnā€™t have ā€œanticipativeā€ processing) better than Cubase? Why ASIO Guard makes everything so much better in Cubase? So many questions and no answers yet :frowning:

No fix for this in 9.5?

Hey guys, just a quick tip as I miraculously found a solution to what seemed an unsolvable problem.
I tried absolutely EVERYTHING mentioned here and nothing seemed to help with my CPU hitting 100% and ASIO dropouts, until I simply turned off (quit) my Kaspersky Internet Securityā€¦the CPU immediately went down to 10-15% and everything is running fine ever since.

Win 10
Intel i7 4770k
Cubase 9.5.30
RME Fireface 400
Waves Soundgrid Server One

Regards,

h3kt0

Thanks for the link.

Same limitation of # of MMCSS threads exist in the latest Windows 10 upgrade 1809.
During playback of a large 400 tracks test Project I get also audio dropouts, although my machine is very nicely tuned for audio.
E5-1650v3 CPU (6 cores, 12 threads).

ive started getting the same problemsā€¦ive had them the past few days I have a Apollo twin usb and it was happily running at a buffer of 126 sice the day I bought it.
but as off sunday if I load up a project I created in Cubase 9.5 it will glitch/slow down/pop unless I raise the buffer too 256 or above even at 256 theres still artifactsā€¦

but I cant understand why when it ran and recorded it fine and was working ok last week at buffers of 126.

the only thing when has changed over the weekend was I bought a Behringer neutron and installed its driver and installed Cubase 9.5.40.i think windows 10 may have updated something too but cant be sure.

my pc has a i7-5820k cpu 3.30ghz/12 logical procesors/32gb ram/windows 10 1083.

any help?

Is there any news on any of this? Have Steinberg ever hinted that theyā€™re working on a way to actually make use of the higher-core-count processors? Not all DAWs are limited in this way, after all.

Havenā€™t read the full thread, so bear with me.

I have really nice even load in Cubase on 8 cores(i7 quadcore with hyperthreading).

But have issues with other stuff in computer that in regular intervals do things that cause crackle at playback, or if recording it stops and give message of dropout.

I have eliminated a bunch of software that do scheduled stuff, like first 20 minutes after boot and similar, various checks for updates even though no network is up etc. But still have an ongoing one - every hour, on 5 seconds distance - that cause this problem still. Repeatable - next boot - take note of when booted up - and within two minutes from that first dropout will appear, and then on 5-10s from that point - all day through.

Spent full day yesterday and still see no immediate cpu increase on any running service or process overall. Gone through all scheduled tasks and nothing. Run auditing to see what processes start and exit - and still nothing.

Next step is to run process explorer and see if any thread in Cubase itself?
I have no autosave on jobs active.

I have this thread a bit below about this - if there is a setting and Cubase it too sensitive to dropouts loosing the sync?

I remember Sonar has such a setting how long dropout should make transport stop.

I will make a test today with Sonar and see if in computer something can disturb audio enough - without obvious jump in cpu.
I filmed task manager in both process explorer and task manager to see when it happends.

Other tips from me
a) Task scheduler in Administrative tools - there you have a good bunch of things that are triggered on various events.
There I found most of disturbances.
b) Security soemthing, also in Administrative tools - auditing can be help to narrow down if dropout occur to a process that is started on scheduled stuff. Look both in Local Audit and Advanced audit to enable those - process tracking, I think it was called.

@Larioso
Nvidia drivers are the worst, check it and try different versions. Latest doesnĀ“t always mean best, as weĀ“ve seen with SteinbergĀ“s update

Thank you, have not updated since in spring some time, but will update and see if any difference.
I removed a full bunch of Nvidia stuff in Task Scheduler, that was already disabled since long - in msconfig Startup.

I tested right now in Sonar - no problem at all, Continuous record and no sign of dropouts.

I sense that Cubase is very sensitive in how it intercept it looses sync?

Have set audio to boost after my testa started, and also Steinberg power scheme - no difference.

About to test setting External sync on - where Cubase allow samplerate to run from another source - will see if any difference.

EDIT: No change on this, external clock become the same
And it comes on the second hourly from the first system event log for the day, about windows starting up.
So something is scheduled then - and create a problem in Cubase - mark not Sonar - only Cubase.

So does the registry fix solve the problem now? Iā€™m planning a new computer build and Iā€™m thinking of i9 9900k (8 core, 16 thread). Would it be totally pointless if we need to disable 2 threads anyway?

To the best of my knowledge, the registry fix makes things more stable but yes youā€™re abandoning anything over 7 cores with hyperthreading. So this new Basin Falls range is largely a waste of time for Cubase users. Really hoping C10 changes something in the Audio Engine to get round thisā€¦ the registry ā€œfixā€ is certainly no long term solution and Steinberg will be making themselves increasingly irrelevant if Cubase canā€™t use new technology.