Windows performance analyser, CPU usage attributed vs precise

Joop · 26 Mar 2023

Dear members,
I am very new in this forum and hopefully this question is in the right place.

I am looking into the performance of my computer (HP EliteDesk 800 mini , i7) that I am using for real time audio (Cantabile).
I am trying to understand for quite some time why I get audio glitches when the internal monitor in Cantabile reports a CPU usage of only 10-20%. I'd hope to understand this better by using Windows Performance Analyser.
The first graph on CPU (CPU usage attributed) shows me that the CPU is used 100%. However in the precise graphs I see a usage of less than 10% (Usage per process). I also see that 1 of the processors is used 100% and the others not at all. Is that a reason that the attributed usage is 100%

I would be happy with any explanation, in particularly wat the attributed usage graphs shows,

cheers,
Joop

dalchina · 26 Mar 2023

I am trying to understand for quite some time why I get audio glitches

Hi, try running LatencyMon (free) as this is a more appropriate tool:

LatencyMon checks if a system running Windows is suitable for processing real-time audio and other tasks

and see what that tells you.

You can find example threads (relatively old now) where Latencymon has been used by searching tenforums for that.

As for the WPA, @zinou is the expert.

Joop · 26 Mar 2023

Hi Dalchina,

I did run LatencyMon, but that is apparently only part of the story
It typically tells me:
"Your system appears to be suitable for handling real-time audio and other tasks without dropouts. "

But it does not tell me why the CPU load of my system is still low when glitches arises. I actually think now that this has to do with the distribution of the load over the processors, but I would like to understand this better and think that WPA can help me with this.
It further tells me that the DPC/ISR bottleneck is:
Wdf01000.sys - Kernel Mode Driver Framework Runtime, Microsoft Corporation
which is not really specific.
@zinou, can you help me with this?

Joop

Malneb · 31 Mar 2023

If by glitching you mean the sound is choppy or dropping out then you need to look at your buffer inside the music software settings.

Maybe some more context because you are not exactly saying what the issue is with the software glitching in this context could mean a few things so i am making an assumption here but generally low CPU usage and glitchy sound in any daw will mean buffer latency or underruns or buffer not big enough etc assuming this is what you mean by glitchy.

Default audio drivers that are for generic audio devices on any computer are meant for playback not processing real time audio and you need a third party sound card for anything outside of that. So if you don't have a proper sound card then you want to use a tailored driver for real time audio called Asio4all

This is a general computer forum you may get more specific answers on their forum instead as your issue is more niche in circumstance.

Joop · 02 Apr 2023

Hi Malneb,

I'm always pleased if somebody takes the time to answer the question. The audio in this is the context but my question is on Windows Performance Analyser. So, I repeat the question:
The first graph on CPU (CPU usage attributed) shows me that the CPU is used 100% (more recently also >100%). However in the precise graphs I see a usage of less than 10%. I also see that 1 of the processors is used 100% and the others not at all. Is that a reason that the attributed usage is 100% and the precise usage is 10%?
I'd also like to understand better the difference between the precise and sampled graphs.

On stackoverflow I found on the precise graphs: "Therefore, CPU Usage (Precise) data knows how much time a thread is using but it has no ideas where that time is spent." My question: what is meant with "where"? Does it mean on which processor?

On audio fora I can find the typical information for real time audio. Buffers etc. is explained there in detail. Understanding multi-core processing is less main stream there, and I hoped to find that knowledge here,

regards,
Joop

Malneb · 02 Apr 2023

Forget Windows Performance Analyzer. What do you expect to achieve exactly by using it? it is meant for software developers to utilize how their software is interacting with the operating system,
so unless you are able to re code this application from source then its not going to serve you any purpose in relation to your issue and personally i feel you are looking in the wrong place tbh.

The parameters you have given so far i would still say it is a buffer issue. Glitchy real time audio only come from one place and you won't be fixing it from WPA.

- - - Updated - - -

Here is an example using the primary sound driver in windows regardless the concept here still applies no matter what audio driver you are using..

The buffer underruns are sporadic with a tiny buffer and the audio is glitchy/choppy, When i start to open up the buffer size it becomes more stable and the audio quality improves and is playable. You can also see up the top near the peak meter that the CPU load is always low even when the sound quality is bad and the buffer is small. The interface is also glitchy but start to improve over time.

CPU load in your software is the slice that is being used by your whole project, CPU Load and Utilization are two different things 100% Utilization is good because it means that the CPU is being used. I guess you need to first establish what you are reading what your issue is and how to solve it because atm you are also looking at the wrong metrics.

Joop · 03 Apr 2023

Hi Malneb,

thanks again. I know that a longer buffer will eliminate glitches. Because I'm a live performer latency should be minimal. Presently I am playing with the Steinberg UR22C, Steinberg/Yamaha ASIO driver with a buffer of 128 sample/low latency. That is something like 64 normal. Mostly that works out, but I want to improve. Can't show you an image now as I'm not at home and do not have the interface here.

Actually you can do something as performer. Typically the DAW (Cantabile in my case) is able to do the multi-core processing for you, and you can set the number of cores that are available for the DAW. In some VST's, e.g. Kontakt player, you can set the number of cores at well. Most manuals advice to give all the cores to the CAW or to the VST's. I'm actually finding now the best results with giving some to the VST (3) and the others to the DAW (9). That gives me good performance at lower buffer size and I see on WPA that the cores are loaded better. So I try to make sense of this.

- - - Updated - - -

Hi Malneb,

thanks again. I know that a longer buffer will eliminate glitches. Because I'm a live performer latency should be minimal. Presently I am playing with the Steinberg UR22C, Steinberg/Yamaha ASIO driver with a buffer of 128 sample/low latency. That is something like 64 normal. Mostly that works out, but I want to improve. Can't show you an image now as I'm not at home and do not have the interface here.

Actually you can do something as performer. Typically the DAW (Cantabile in my case) is able to do the multi-core processing for you, and you can set the number of cores that are available for the DAW. In some VST's, e.g. Kontakt player, you can set the number of cores at well. Most manuals advice to give all the cores to the CAW or to the VST's. I'm actually finding now the best results with giving some to the VST (3) and the others to the DAW (9). That gives me good performance at lower buffer size and I see on WPA that the cores are loaded better. So I try to make sense of this.

Malneb · 04 Apr 2023

well honestly 128/44.1 assuming that is your sample rate no real reason why it shouldn't be is going to be 2.9ms in latency and imo not noticeable.
1000 ms is 1 second so unless you the flash and breaking the sound barrier when you run then its not really going to make a big difference people say they think they can feel better response but in the milliseconds we are talking and in the whole scope of it is not really that marginal.

- - - Updated - - -

i guess look at this scale of things that happen sub 1 second or sub 1000ms we measuring bees dicks at this point
Millisecond - Wikipedia

in your case 144/44.1 2.9ms latency would sit here on the chart 3 milliseconds – a housefly's wing flap. Also the normative speed of sound.

live performance sure you want this sort of latency but if you cannot really achieve it because of any several factors its not a big deal. Most of the audience like 95% of them is not going to know anyway and to really differentiate lag to yourself while you play idk man some people say they can notice some can't.

256/44.1 would be 6ms and still perfectly fine like 6ms man is 3ms or 0.003 seconds really that noticeable? or else you know what comes next you need to compromise somewhere because you are filling up the buffer to much its one or the other can't always have both this is just generalization here because there could be several things at play without knowing all the parameters but you know what i mean.

- - - Updated - - -

I guess the next step is are you totally DAW bound or are you actually processing live instruments?

I have heard of Cantabile out there but not overly familiar with it as in never used it so i did some research on the software, it is interesting to note that they are of the same opinion.

Excerpt from one of their pdf.

Sample Rate and Buffer Size

In the Introduction to Digital Audiochapter at the start of this book I described how we can use the limits of human anatomy
to determine a reasonable buffer size and sample rate for good quality audio.

We determined that anything humans can hear can be accurately represented with a sample rate of 44,100Hz and a bit depth of 16-bit.
We also determined that humans can’t discern latencies less than about 10ms and that a buffer size of 256 samples will satisfy that limit

providing latency of about 6ms.

This of course changes if you are doing live instruments but there would be to many factors involved and i would have to check out in this case as only you really know what is happening. Regardless you know what to do.

- - - Updated - - -

You also have to realize the majority of DAW software out there are at the pinnacle understanding of how to interface a computer in real time there is not much room outside of this as a user to improve upon this because they know how to design the software its common to see whitepapers and in depth study about these things.

Its all down to whatever you are doing as a user and what you are trying to archive. I can play in my DAW at 1ms with no visible latency and no audio degradation if i am using the right adapter but that is not achieving me much is it? Except for trying to push a whole lot of data through a small hole but at the same time i don't play live instruments.

- - - Updated - - -

You state being a live performer so this can mean a few things anything from being DAW bound like playing VST and effect through midi is still real time audio or processing actual chains of live instruments in real time which is far more complicated.

You never really specified and the latter is more complicated. if you are just like DJ or something its far less crucial.

in case of live instruments then you should probably ask in more specific forums at the end of the day i don't think there is much to be had by trying to analyze what the DAW is doing because its doing its job. Then again you would probably know what you are doing if live instruments at the same time and not really ask these sorts of questions.

Joop · 04 Apr 2023

Hi Malneb,
yes 256 will probably do, so I tend to use 128 to be on the safe side, but beside of your essay there is still the question how to assign the cores and there the opinions are not unanimous. In the end I will stretch the application until the limit and that happened already. It make sense to move up the limit a little bit.
By the way I just have keys and all my instruments come from my DAW.

Malneb · 04 Apr 2023

Yeah an essay you should actually try reading and you did also ask for help in forum which is a place where people speak their opinions so don't try to insult my intelligence by having an attitude thanks, i am trying to help you at the end of the day.

128 is not on the safe side a higher buffer is safer not a smaller one by this statement alone i think you must be new to all this.
a small buffer is not being safe its the total opposite and running with no head room while trying to squeeze out latency but this also means that there is lots to go wrong and if your audio is glitchy then you need to increase your buffer.

You are just playing vst with midi so such a small buffer is not that big of a deal just aim for 10ms if that does not work go higher until you actually know what you are doing and then you can start worry about small buffers.
You could do 512/44.1 which is 12ms perfectly fine for your situation even 1024/44.1 at 24ms anything like this is fine and the top priority in your case being that you are only doing midi and essentially DAW bound is sound quality not latency.

Latency is for recording audio or playing analog instruments mainly. There will be some lag on high buffer with midi but again its not that big of a deal.

So again in another way as you don't seem to get it if you are playing so many samples, effect and VST at one time that your DAW starts glitching then your priority is to increase the buffer because sound quality comes over latency this the general rule of thumb anyway regardless of situation.

Bottom line its your buffer and probably a combination of using the wrong driver or wrong audio adapter. The application is multi threaded and you are looking at the wrong metrics, so if it is only pegging one core then it will be due to two things either Hyper threading is disabled in your bios or you don't have the software setup properly.

it is also strange that this is only coming out now and your initial complaint was about low cpu load/ low cpu utilization you never once said anything about core count tbh i am not buying it and i don't think you really know what your problem is and how to solve it i also think you may be slightly miss informed in your logic or chasing a red hearing.

You also might want to look over this if you have not and understand the chain of events to help you diagnose your issue.
Understanding Cantabile'''s Multi-processor Support - Cantabile - Software for Performing Musicians

the software should also have some sort of shutdown feature where it turns off stuff dynamically when not being used make sure this is enabled.

crawl before you walk.