Triggered by some statements about KIT Plugin's BB N105 I duplicated a track with this plugin several times and found out that the CPU load per plugin increased! From 1,7% to 2,7% at 35 tracks, which resulted in a total CPU load of 94,5%. But at 1,7% it would've been possible to run 55 tracks with that total CPU load.

I found the same behaviour for the Weiss EQ and AA Cerise. Millennia's NSEQ-2 is not doing this.

Does anyone have an explanation for this?

BTW, Win7, Reaper, just in case.

 

Hmmm...
If you insert several instances spread over several tracks, and turn the preamps up the same amount, do all instances deliver the same measurement results in the spectrum?

Or maybe the plugin has problems with multi-thread queries.

 

Last edited: Jun 26, 2021

The plugin itself is nice indeed, but do you reall need to run 55 instances?

that being said, the plugin is still practically new, maybe new updates will bring cpu optimisations ?

 

No clue, I tried it with a random source (drums).

EDIT: But interesting idea. I just tried with a sine wave, phase inverted the first copy and duplicated this pair. Flatline even with 30 tracks.

Nope. Since 35 already caused 94,5% CPU load, it wouldn't work anyways. But that's not the point (as you know). I find this behaviour strange.

 

Last edited: Jun 26, 2021

You don't say?

 

I asked, because there are some plugins that communicate with each other. Had no chance to test the plugin, so I do not know if this is the case here at all. With Cerise I have not noticed this behavior, related to CPU consumption.

 

I will post the link below the screenshot, but this is at the root of why, though there are other considerations - look for more on FLS: (Win 7 + Win 8 have a basic FLS, but it is not as extended as it is in Win 10 1903 onwards)




LINK:

https://devblogs.microsoft.com/wind...-the-fls-slot-allocation-limit-in-windows-10/

 

Ah, you missed my EDIT by a minute

 

have you noticed that some developers take multiple instances completely into consideration and others do not?
For example, if the plugin has a large graphic load, the plugin detects that it has already been loaded once and does not perpetually reload the graphics as separate instances. While GPU is not often taken into consideration, if it crashes, so does the DAW and often the O/S. A decent GPU in a DAW is always a good thing.
Another is it detects that because it has a grouping option, or a load requirement, it splits the load mong multiple instances instead of a fresh allocation every time.
Unfortunately, not all developers think that far ahead.

 

Holy shit, if anyone is Acustica. It's this shit.

 

Last edited: Jun 26, 2021

This happen to me as well with some plugins. With Acustica Audio is really evident, each instance of Cream 2 has different cpu usage having the same settings and the more instances you load the more cpu each plugin uses, now testing with the Millennia's NSEQ-2 and the cpu usage stay normal regardless of the number of instances. I asked something similar while ago thinking that this problem was exclusive of Studio One without a clear answer
https://audiosex.pro/threads/studio-one-4-6-cpu-problem.58479/

 

If anyone REALLY wants to bench test the "How much can I load before crashing"

Try 5 instances of Superior drummer 3 and then 5 instances of Spectrasonics Omnisphere.
If your system has not fallen over in a screaming heap great!! Now add Kontakt and add 5 sample libraries into a single instance.

If you have the Lurssen Mastering Console, now add that on your stereo bus. If the audio sounds like a fart through an AM radio then your system is on the edge of falling over. If you have done all of the above and the audio sounds clean, then you probably have as stable a DAW system as anyone could hope for, even though your CPU usage, RAM usage and disk caching may well be sitting around 95% non-stop at the very best.

 

Absolutely normal.
The CPU load is not related with the action of processing a buffer BEFORE a deadline.
It is particularly evident if the plugin consumes resources (not necessarily the CPU but also memory and cache bandwidth).
Basically your examples.

Just to explain in an easy way: you could have a process taking just 10% of your CPU in a 10 core computer but can't be scaled or it just needs time for completing. Basically it is consuming 100% of a core, so 100% of the time.

The computer is barely managing it and the asio meter is 100%, and you have your computer clicking and crackling, but apparently your task manager tells you there is 90% CPU free.

The situation is even more complex if process are distributed across cores. Maybe your plugin is 10% but it is distributed across cores and it needs time for completing. The core time slices are occupied, so it is hard for multiple plugins to find the same resources free in the time slots they need, and their allocation is less efficient than the first one.

 

Last edited: Jun 26, 2021

If you already use VST3 there is not much you can do but increase ASIO Buffers or try more VST3 versions of your plugins.

Copy pasted of "https://thelittleredlight.com/gear-reviews/vst-vs-vst3-difference/"

"The biggest advance with VST3 is that it only applies processing if there is an audio signal present. This reduces that weight on the computer’s CPU, allowing greatly increased performance."

"Improved performance and lighter on the CPU. It only applies processing when there is an audio signal present, freeing up resources"

Years ago I started to make sure i use the VST3 versions and it indeed helped a lot on my old Xeon-1230v3 because there are never 100% of my plugins are used at the same time.

Increasing my ASIO Buffers from 128-256 to 512+ also helps finishing heavy loaded tracks if the Latency sensitive work has been finished.

 

It's not about that at all.

 

So why is NSEQ-2 behaving differently?

And ecxatly this is not the case.

1 track = 1,7%
24 tracks ~ 2,6 to 3,3%. In sum ~ 68% in Reaper and in the task manager, for every core/thread, all with the same load. No spike even close to 90%.

So, while everything's running smoothly, no drop outs, no core/thread overloaded, every instance uses 53 - 94% more than it does with only eight or ten instances.

Some plugins behave like this and some not. That's why I'm asking.

 

BB N105 is VST3 only. Other than that, I try to avoid these versions.

It's not about the amount of plugins I can use, it's only about their strange CPU load behaviour.

 

Hi @No Avenger not target-oriented but still worth knowing.
CPU Utilization is Wrong www.brendangregg.com/blog/2017-05-09/cpu-utilization-is-wrong.html