CPU usage vs Cubase performance meter - different

Why?
and how on earth to fix?
__


__
In the attached pic, you see the Cubase performance meter - the green line, like at 80% or so…
…and in the upper right corner you see a small black rectangle with several tiny blue vertical lines that represent all 24 of my CPU threads (12 cores x 2 threads each), with SO MUCH room left to process stuff.

Mac Pro 2010. 64GBs RAM. 512 buffer.

I just don’t get it. Please advise.

The Cubase Performance meter is ASIO performance, not CPU performance, so they aren’t really related. Many articles on this subject.
Here is one.

1 Like

I guess Jaslan beat me to this but …
The Cubase performance meter shows the audio processing load and the hard disk transfer rate. It has more to do with your asio driver and data transfer rate than the CPU usage. If your CPU usage is low and your performance meter is reading high, then you have a bottleneck somewhere else in the system. Maybe your hard disk isn’t supplying data fast enough. You should also check your audio driver. Make sure you have the latest update for it and check your settings to handle multitrack scenarios. Make sure your buffer size isn’t set too low. You can also activate asio guard.

1 Like

Thanks. I appreciate the feedback.
Without having read the articles, and I’m going to read them here in a bit, I just want to give a quick comment…

That bottleneck… hmm… The hard disk meter isn’t reading anything in that pic, I’m running an NVMe drive and my Samples disk and they both stream you know over 300 MB/s or something much higher. All I have is IK Multimedia’s Modo bass, 2 audio files, and the UVI workstation sampler playing one patch. So that is a really light load, which IS reflected in my CPU meters not moving much at all and that hard disk meter within Cubase showing zero.

So I can’t even fathom what could possibly be bottlenecking and causing an 80% ASIO performance spike… Except something to do with ASIO.

64 GB of RAM, so it’s not the RAM amount, is it the RAM speed? Unlikely. It would more likely be the hard drive than the RAM and it’s not the hard drive. It’s not the CPU.

So ASIO drivers… On the Mac, I think most of this stuff is plug and play, so it’s worth double-checking the manufacturer’s website to see if there’s other driver I can install.

Also, ASIO guard. No matter how much I’ve studied exactly what that is, I never fully get it. It seems to only impact performance on the track I’m recording on, right? Like, I can set my project buffer to be something quite high, but with ASIO Guard enabled, I can still record a MIDI part and not experience much latency. Is that it? And if that is, that doesn’t seem like it would affect my overall ASIO performance, but challenge it more.

Anyway, off to do some reading. Something is clearly wrong because that discrepancy between the ASIO meter and the CPU meter is just too big. I will report back.

1 Like

I don’t use a Mac so can’t comment on whether there are just standard drivers built in but I do know that the audio interface and its drivers have a massive effect on the Cubase performance meter on a PC. My RME interface is by far the best of the interfaces I have used and that is, I understand, mainly because they custom wrote their own ASIO drivers.

Thanks. According to one of the commenters (in the thread that’s linked to in the first reply to this post above, by jaslan), that video only tells half the story and gets some things wrong. It’s worth a read. (The commenter’s name is Glen0 or something.)

Yeah, and I guess also VSTs, how they’re written for real-time performance and offline performance, and how much you have going on on one track.

The posters above are correct in saying the VST Performance is not the same as the system’s CPU usage.
VST Performance measures the CPU load within the constraints of the ASIO driver, a CPU meter does not.

Also, if you have a single core which is working hard, the VST Performance meter will show its value, not an average of all the cores (due to the fact that if one maxes out a single core, the application will stutter, regardless of how low the load is on the other cores).
This in turn means that if you have a small project with just a few, say ‘X’, VST Instrument tracks with FX processing on them, you’ll be using ‘X’ amount of cores (‘X’ being 1/4 of the available cores) - if you now duplicate those tracks, the additional instances will be allocated to different threads, previously unused, the result being little to no increase in the VST Performance.
Kind of an edge-case and indeed a simplified example, but helps understand the mechanism.

ASIO Guard is designed to use virtual threads: if you disable it, you effectively use the physical cores only.
Tracks being recorded can use ASIO Guard, if monitoring is disabled.
The current engine uses physical cores for time-critical processing and virtual cores for ASIO Guard / non-time-critical processing.

As sometimes an image says more than a thousand words, in the attachments you can see: 20 seconds (2 leftmost squares) with no tracks selected, 20 seconds with all tracks recording (no monitoring enabled) and 20 seconds with all of the tracks having monitoring enabled.

Thanks Fabio. Given my specific scenario, 2 audio files, 2 VSTis, a couple of plugins, compared to the relative power of my system, it would seem then that I am somehow over-taxing the ASIO component on just these particular individual tracks, right?

So basically, my Cubase VST performance meter that’s showing 20% overhead is only showing the overhead for THOSE tracks, it’s not actually showing the correct overhead for adding new tracks. For new tracks, I presume I have a 100% overhead, at least until I fill up all my cores and threads with stuff.

But if that’s the case, wouldn’t the CPU cores or threads that I AM currently using show 80% utilization, like what my ASIO performance meter is telling me? Because my CPU performance on all cores is barely moving, as far as I can tell.

You’re basically just loading part of your resources, cores in this case, and the load you see in VST Performance is that of your busiest thread (while a few of them just sit watching a couple do all the work :smiley: ).
While you keep adding tracks and effects, it gets closer to showing what you are actually left with in total, as all cores get progressively ‘filled up’.

It might be worth mentioning that the cost of thread synchronisation is very high with light loads, while with high loads actual processing becomes the predominant part.
In short: with such a small project, you might actually see a lower load with HT disabled. Not recommending to disable it, just mentioning it.
If you keep on build up on that project, you might see very small variations while you keep on adding.

The OS CPU meter does not show 80% because as discussed it does not take into consideration the driver constraints, latency above all.
If you were to use a high latency, then the VST Performace and the CPU performance should show similar, or at least close enough values - the more you lower the latency, the higher VST will be compared to CPU.

As Fabio has pointed out, there isn’t necessarily anything unusual in what you’re observing. I’ll just mention that the Performance window often gives more insight into situations like this than the info line cpu meter you’re looking at. Also, temporarily turning off asio guard sometimes helps one make more sense of what the meter is showing. Also, the plugins you’re using may be constrained by resources other than cpu. Finally, I’ll mention, especially with asio guard and multicore processing, the performance meter doesn’t really show a measure of capacity utilization. Back in the days of single core processors, that meter was a simple thing to interpret, but that’s no longer the case :slight_smile:.

Thank you, gents. Makes much more sense.

It’s a relief knowing that my system doesn’t need some major overhaul.

This page is now bookmarked. I’m sure I’ll have to refer back.

Thinking through this, computer performance is like a group of people working on a project - where everyone is working at 100% capacity, but some people have more to do. So while those people are working to finish up their bits, the other folks have to sit around and twiddle their thumbs, doing nothing. So there’s no way to max out this entire group of people to 100% efficiency… UNLESS, of course, everyone helps each other out at ALL times… which, as we’ve all learned at some point or another, is massively inefficient. You lose time explaining what you’re doing to people, and you just get frustrated.

But if anything could eventually do this, it would be a computer… I guess.

Still, the act of coordinating moving parts seems like it will always have limits - always something waiting for something else.

Huh… I guess I’ll go off now and have a good cry that my computer will never EVER show 100% CPU utilization. :frowning:
…except for that dang VST Performance meter. :slight_smile:

We’ve sent so many views to this vid over the years…lol :laughing:

Or something :slight_smile:.

Same here. I have the Asus Proart z490 proart mobo, 2 nvme drives and intel i9 10850k processor with 128g RAM and my files from Cubase 10.5 on my laptop with an i7 chip and 32 gigs of RAM are choking on my new Desktop supermachine. Iy even chokes with much smaller files. I am baffled

I found this article “if you have more than 7 cores on your CPU then ASIO-Guard fails when you arm too many tracks and some VST instrument tracks are included in those armed tracks” https://www.reddit.com/r/cubase/comments/ilegxh/105_one_major_issue_asioguard_multicore_support/