The No.1 Website for Pro Audio
Here it is: what actually impacts performance in Cubase
Old 22nd May 2020
  #1
Here it is: what actually impacts performance in Cubase

A reoccurring topic in the Cubase forum is performance problems. Here is a thread for sharing confirmed, generally applicable, tips and tricks.

Obviously, your CPU and the audio interface drivers have a huge impact. Those are discussed in the two great pinned threads in the Music Computers forum. So this thread is for stuff we can influence apart from those factors.

Please do not post anecdotal, hearsay, "I read it on the internet", type hacks. For operating system level tips, only post officially supported or very established stuff.

I will start out. First the basics, then some carefully documented findings on how Cubase works. Let me know if there are other use cases that i can run through my setup.

Hope it's useful

What are your tips, tricks and learnings for great Cubase perfomance?

Last edited by thedberg; 22nd May 2020 at 07:56 PM..
Old 22nd May 2020
  #2
Ultimate Perfomance mode

If you're on Windows 10, this is, a far as I know, the only really valuable setting/hack. You really don't want your computer to go into sleep mode in the middle of a long recording stint.

Here is an official MS post about it:
https://answers.microsoft.com/en-us/...0-35197275d083
Old 22nd May 2020
  #3
Latency mon

Another thing you really don't want is to have processes running in the background that kidnaps the CPU for longer than the allowed time. That will result in dropouts. Find those buggers in Win 10 with LatencyMon.

https://www.resplendence.com/latencymon

See the thread Today We Build Or Studio PC for help on how to interpret the results.
Old 22nd May 2020
  #4
Cubase can balance cores well in playback

Now comes the meat and potatoes of this thread: how Cubase works.

First off, a description of my computer.
  • AMD Ryzen 3900X 12 core CPU, no overclocking
  • Computer from Scan audio, setup for DAW use
  • No other disturbing software (no games and ****...)
  • RME HDSPe pcie interface
  • Windows 10 professional, no tweaks other than Ultimate Performance
This maybe a bit more powerful than many of the computers in use here. But as it's totally focused on audio, I think the results are generalizable and applicable for many.

And here's the setup in Cubase for most tests
  • Cubase Pro, 10.5.12
  • 64 sample buffer
  • 48 Massive X tracks, all burdened with S-Gear and and TDR Kotelnikov GE plugins
  • 48 audio tracks with Kotelnikov
  • ASIO guard: High
I chose those plugins because they're some of the most demanding I own. And no, I don't own Acustica stuff, it's incompatible with my workflow.

It's obviously a contrived setup, no real project is so easily balanced by Cubase. But as we will see, it can teach us some important lessons.

When playing back, my computer acts like in the attached pictures.

CONCLUSION
With ASIO Guard High Cubase can scale very well to multiple cores when faced with a load that allows it.
Attached Thumbnails
Here it is: what actually impacts performance in Cubase-cpu-playback.png   Here it is: what actually impacts performance in Cubase-perf-playback.png  

Last edited by thedberg; 22nd May 2020 at 07:32 PM..
Old 22nd May 2020
  #5
Recording

With 8 of the tracks Recording, Cubase acts as in the attachments. ASIO Guard still High.

Note that
  • Balancing between cores is slighty more uneven
  • There are occasional peaks on the ASIO meter
  • Real-time peaks are on a radically higher level

CONCLUSION
Be prepared that recording, or even arming tracks for monitoring, makes performance worse.
Attached Thumbnails
Here it is: what actually impacts performance in Cubase-perf-record.png   Here it is: what actually impacts performance in Cubase-cpu-record.png  
Old 22nd May 2020
  #6
Group processing

Now it gets more interesting. What happens when we pass audio through a single bus/group?

I've now added several TDR Limiter 6 to the master bus (heavy stuff, both CPU and sound wise). This is to mimic a typical case when we load a single group or bus with plugins that put a lot of demand on the CPU.

This give results that can be listened to but with a lot of noticable dropuots.

I have done the same test but on a group bus before the master. Predictively, Cubase behaves exactly the same with group processing as with master processing.

Note that
  • One core (third from the top) is much more loaded
  • Total CPU utilization goes down, because most cores spend time idle, waiting for that core to finish

CONCLUSION
A single track, bus or group needs a single core, so avoid putting a lot of processing in the same serial chain if you get performance problems.

Torbjörn's trick
My guess is that this is a common source of performance issues. My trick is to avoid heavy processing on the master until late in the project. At that stage I'm more into mixing than recording so I can set the buffer higher. This is also good practice because I don't rely on magical stuff on the master to sound good.

Obviously, this is only applicable in some use cases/workflows.

What is your trick to handle this situation?
Attached Thumbnails
Here it is: what actually impacts performance in Cubase-perf-group.png   Here it is: what actually impacts performance in Cubase-cpu-group.png  

Last edited by thedberg; 22nd May 2020 at 07:38 PM..
Old 22nd May 2020
  #7
ASIO Guard Off

As far as I understand, ASIO Guard High adds a large (4096 sample) buffer for non-realtime processing (playback). So what happens when we turn it off?

To get listenable audio, I have to reduce the number of track in my setup from 96 to 48.

Note that
  • Core balancing is totally off
  • Realtime peaks abound (also heard)
But:
Putting tracks in Record makes no difference in ASIO handling or CPU load.

CONCLUSION
ASIO Guard Off screws up load balancing completely and worsens performance a lot. But recording doesn't impact CPU load like it does with ASIO Guard On. Stable recording of multiple sources may possibly be a use case for having ASIO Guard off.
Attached Thumbnails
Here it is: what actually impacts performance in Cubase-perf-off.png   Here it is: what actually impacts performance in Cubase-cpu-off.png  
Old 23rd May 2020
  #8
Lives for gear
 
DeadPoet's Avatar
Hi,

Thanks for doing this - I've dived in this pool before as well and always happy to see people trying to understand the way Cubendo deals with audio and CPU horsepower.

I've downloaded the Kotelnikov plug (non GE) and I do not have the S-Gear plugin.


Win10 latest, CP10.5.12, 3950x, 64Gb at 3600, RME madi-card.


*) Loading 64x MassiveX+Kotelnikov and 64x mono audiotrack with Kotelnikov at 64 buffer size with ASIO normal.
Gives me a stable asio meter just below 50% and real-time peak just below 10%. TaskMgr showing even load of about 43%.

*) recording 8 tracks makes no noticeable difference here.

*) switched my asio guard to high, re-did above stuff and that makes no difference.
*) switched asio guard to low and CPU jumped a bit to just over 50% on the ASIO meter (49% in TaskMgr)

*) disabled asio guard and everything's in the red. CPU at 80%. There's no audio in the project but it looks unplayable.

*) when I load 8x Kotelnikov on the "Stereo Out" the ASIO jumps to about 70% (real-time at around 50%), TaskMgr at 51% and all even-numbered cores are now higher in cpu load than the odd-numbered

*) when I insert a stereo bus between all the tracks and the StereoOut and load up all the kotelnikovs there I get asio at 75%, real-time peak at about 4-5% and TaskMgr at 46%
This confirms a test I did a long-time ago: you get more mileage out of your system if you do not load plugins on the Output Busses of Cubase but use a "pre-master" buss instead.



Hope this helps and I'm curious to read other people's experience on the matter


Herwig
Old 23rd May 2020
  #9
Quote:
Originally Posted by DeadPoet View Post
Hi,

Thanks for doing this - I've dived in this pool before as well and always happy to see people trying to understand the way Cubendo deals with audio and CPU horsepower.

I've downloaded the Kotelnikov plug (non GE) and I do not have the S-Gear plugin.


Win10 latest, CP10.5.12, 3950x, 64Gb at 3600, RME madi-card.


*) Loading 64x MassiveX+Kotelnikov and 64x mono audiotrack with Kotelnikov at 64 buffer size with ASIO normal.
Gives me a stable asio meter just below 50% and real-time peak just below 10%. TaskMgr showing even load of about 43%.

*) recording 8 tracks makes no noticeable difference here.

*) switched my asio guard to high, re-did above stuff and that makes no difference.
*) switched asio guard to low and CPU jumped a bit to just over 50% on the ASIO meter (49% in TaskMgr)

*) disabled asio guard and everything's in the red. CPU at 80%. There's no audio in the project but it looks unplayable.

*) when I load 8x Kotelnikov on the "Stereo Out" the ASIO jumps to about 70% (real-time at around 50%), TaskMgr at 51% and all even-numbered cores are now higher in cpu load than the odd-numbered

*) when I insert a stereo bus between all the tracks and the StereoOut and load up all the kotelnikovs there I get asio at 75%, real-time peak at about 4-5% and TaskMgr at 46%
This confirms a test I did a long-time ago: you get more mileage out of your system if you do not load plugins on the Output Busses of Cubase but use a "pre-master" buss instead.



Hope this helps and I'm curious to read other people's experience on the matter


Herwig
Great! You went to the effort of redoing the same tests. Fantastic

If I understand correctly we got similar results except for:
• You noticed no difference when recording with ASIO Guard on. I wonder why - the computer should reasonably have some extra work to do when being forced to process some tracks in real time. Very interesting.
• You did notice a difference when using a group vs using the master bus. That’s also interesting and I don’t understand why there is such a difference; it should be the same amount of work. Or does the master bus have other duties that we don’t know about...

I wonder what other people’s experiences are.
Old 23rd May 2020
  #10
Lives for gear
 
schmuck's Avatar
Thanks a lot for the info (love the CONCLUSION bits)!

I record high track numbers at low latency, and asio guard definitively generates clicks & pops. I only switch it on later in the mixing process, when I start using plugs.

CONCLUSION: keep asio guard off for recording many audio channels at low latency


GROUP tracks: you could cascade several of them and only have one or two plugs on each (I had no idea about one core per track - especially when using several limiters with high oversampling, this could help a lot!)
Old 23rd May 2020
  #11
Group vs master bus processing

Here are some findings that differ from @ DeadPoet 's.

To my setup I
  • added five groups in series that all 96 tracks go through with one heavy compressor (Kotelnikov) on each
  • added five of the same compressors with the same setting in series on the master bus

As you can see from the screenshots, the perfomance seems identical when switching between the two variants.

MY CONCLUSION
Serial group processing makes no difference compared to serial processing on the master bus. There are no performance gains using groups vs master.

I am, however, happy to be proven wrong if my testing isn't good or if there are other scenarios with different behaviour
Attached Thumbnails
Here it is: what actually impacts performance in Cubase-perf-5-groups.png   Here it is: what actually impacts performance in Cubase-cpu-5-groups.png   Here it is: what actually impacts performance in Cubase-perf-5-kotelnikovs-master.png   Here it is: what actually impacts performance in Cubase-cpu-5-kotelnikovs-master.png  

Last edited by thedberg; 24th May 2020 at 12:27 PM..
Old 4 weeks ago
  #12
Here for the gear
 

Quote:
Originally Posted by thedberg View Post
If you're on Windows 10, this is, a far as I know, the only really valuable setting/hack. You really don't want your computer to go into sleep mode in the middle of a long recording stint.

Here is an official MS post about it:
https://answers.microsoft.com/en-us/...0-35197275d083
At least with newer versions, cubase installs it’s own windows power profile which it uses unless you deactivate it in the preferences. Includes all the typical settings as far as I’ve seen.
Old 4 weeks ago
  #13
Here for the gear
 

When I bought a new computer two years ago (i7-8700k, 6 cores) I did some tests with hyperthreading on or off. I used the DAWbench project and one project of mine.
Interestingly, my results showed that ASIO load was lower with hyperthreading off, not extremely, but recognizably so. I think the official Steinberg recommendation is on.
This might very well depend on the kind of project you have. I imagine if you work with a lot of audio tracks or massive sample libraries that HT on works better, as HT is most efficient with compute loads that often wait for resources like I/O.
For workloads where pure DSP calculations (e.g. demanding softsynths or FX) are in the majority, this might be different.
I’ll advise to make your own tests, though, or ignore it and just make music
Old 4 weeks ago
  #14
Quote:
Originally Posted by Herzton View Post
When I bought a new computer two years ago (i7-8700k, 6 cores) I did some tests with hyperthreading on or off. I used the DAWbench project and one project of mine.
Interestingly, my results showed that ASIO load was lower with hyperthreading off, not extremely, but recognizably so. I think the official Steinberg recommendation is on.
This might very well depend on the kind of project you have. I imagine if you work with a lot of audio tracks or massive sample libraries that HT on works better, as HT is most efficient with compute loads that often wait for resources like I/O.
For workloads where pure DSP calculations (e.g. demanding softsynths or FX) are in the majority, this might be different.
I’ll advise to make your own tests, though, or ignore it and just make music
That doesn’t seem unreasonable.

Speculation: Loads that are easy to multi thread, like my test project above, could fare well from hyper threading. Loads that are hard to multi thread, like projects with very different cpu demands on different tracks for example heavy processing on the master, might have more use for dedicated cores (hyper threading off).

Could be tested, as you said, but I won’t do it now.
📝 Reply
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
🖨️ Show Printable Version
✉️ Email this Page
🔍 Search thread
♾️ Similar Threads
🎙️ View mentioned gear