CPU Utilization is Wrong

[AresKrieger]

Well that's why  most of us use things like HWinfo64 and ignore taskmanager, but whatever I guess this was more so netflix trying to relay the message to normies (not that normies will notice or care)

[TimeOmnivore]

6 hours ago, jman629 said:

Why would this matter for Netflix? Well Netflix uses it for its to adjust video quality automatically "auto scaling rules"

If only Netflix would implement useful features such as changing video quality manually, playback speed, (and fixing their terrible UI), then CPU utilization being wrong wouldn't really matter.

[Mira Yurizaki]

31 minutes ago, AresKrieger said:

Well that's why  most of us use things like HWinfo64 and ignore taskmanager

My question is what is it using to measure utilization? I did see a post from their developer showing some proof that HWinfo is more accurate than Task Manager on Windows 10, but I don't feel satisfied in leaving it at that without knowing what they do.

 

Then again I'm making an assumption that Task Manager measures CPU utilization solely based on System Idle Process scheduling.

[AresKrieger]

4 minutes ago, M.Yurizaki said:

snip

Well I'm not actually sure what it uses tbh, what I can say is that the numbers differ quite a bit from taskmanager so eh, but I also don't use windows 10 and expect its measurements to be even worse

[Mira Yurizaki]

10 minutes ago, AresKrieger said:

Well I'm not actually sure what it uses tbh, what I can say is that the numbers differ quite a bit from taskmanager so eh, but I also don't use windows 10 and expect its measurements to be even worse

I'm thinking that maybe Task Manager doesn't poll as often before refreshing. i.e., it refreshes every second, but the metric it uses to gather data is different than HWInfo. Like if it polls less often, I would see that providing less accurate results.

 

However I came to realize something else: Task Manager should not be used as an accurate hardware profiling tool anyway. That's not it's purpose, and I have no reason to believe Microsoft either designs it that way or ever has plans to. It's a full blown application and having a hardware profiler with a GUI is asking for performance skewing. An accurate profiler should be as minimally invasive as possible.

[leadeater]

CPU utilization has never been a metric for actual CPU performance utilization, it's a time slice usage metric as in how often is it doing something, anything. An application thread could be using a CPU thread time wise 100% but be doing actually nothing, hence why HT/SMT was such a nice thing when it got introduced and applications back then rather often were spin waiting for resources.

[captain_to_fire]

8 hours ago, Technicolors said:

what's the cpu utilization reporting like in macos? 

Activity Monitor 

https://support.apple.com/en-ph/HT201464 

[DrMacintosh]

9 hours ago, BuckGup said:

and get rid of DRM

Uhh, no. 

[DrMacintosh]

8 hours ago, BuckGup said:

and most things are 21:9 anyways

No they aren’t, like at all. 

[BuckGup]

2 hours ago, DrMacintosh said:

No they aren’t, like at all. 

Most movies are widescreen and I have seen a couple documentaries too

[Notional]

Wow. Spectre and Meltdown. The little gifts from Intel, that just keeps giving and giving. Well, more like taking and taking performance. Well, I finally got my z97 bios update, so I won't suffer from the security holes. Now I get to use half my CPU.

[DrMacintosh]

1 hour ago, BuckGup said:

Most movies are widescreen and I have seen a couple documentaries too

Well movies sure, but not everyday content. That is still dominated by 16:9

[LAwLz]

6 hours ago, Notional said:

Wow. Spectre and Meltdown. The little gifts from Intel, that just keeps giving and giving. Well, more like taking and taking performance. Well, I finally got my z97 bios update, so I won't suffer from the security holes. Now I get to use half my CPU.

Why are yoy saying it's "the little gift from Intel"? It was discovered by Google, and the issues are not unique to Intel. 

 

And no, it's not that you only get to use half of the CPU. The scheduling works fine. It's just that it is reported inaccurately in task manager. But task manager (and utilization monitoring programs in general) have always been inaccurate. It's just that they are slightly more inaccurate now. 

[porina]

I've relatively minor gripes about CPU usage reporting in Windows in general. For example, say you have a single thread load that maxes itself out, the % is basically 1/max_threads of the CPU. I think it would be more useful (to me) if 100% = 1 thread maxed out.

 

Actually, having both could help in different ways. For systems with HT/SMT, if you see it go above 50% overall, you know you've effectively exhausted the cores and are in the HT/SMT zone.

 

Not used Linux in a long time, nor was I that deep into it, the load on that is kinda demand based isn't it? If you have 10 threads wanting attention, it reports 10? That could be another interesting metric, kinda, what would it take to be relatively unlimited in a situation.

[leadeater]

27 minutes ago, porina said:

I've relatively minor gripes about CPU usage reporting in Windows in general. For example, say you have a single thread load that maxes itself out, the % is basically 1/max_threads of the CPU. I think it would be more useful (to me) if 100% = 1 thread maxed out.

Just change Task Manager to display all threads rather than summarized percentage?

 

[jagdtigger]

10 hours ago, DrMacintosh said:

Uhh, no. 

Actually you are wrong, no matter what they do hackers and downloaders wont disappear magically. Not when they use an utterly flawed system which only punish the paying customer base. Funny how they still didnt realize that their idea about DRM is a contradiction.....

[porina]

1 hour ago, leadeater said:

Just change Task Manager to display all threads rather than summarized percentage?

I do that as a matter of routine, and also enable the "show kernel times" option in menu also. But Windows does what I've nicknamed core smear where a single program thread doesn't sit still, and tends to occupy partially both threads of a core. If there is much of that going on, with multiple software, it can become difficult to tell what is where. 

 

On that note, it might also be interesting if it could offer a hybrid view of top processes using CPU time (minor ones are unimportant), and also colour code them so you can see on which core(s) it is running, potentially allowing manual optimisations via affinity. This is beyond Task Manager. Resource Monitor is closer but not quite there.

[leadeater]

9 minutes ago, porina said:

I do that as a matter of routine, and also enable the "show kernel times" option in menu also. But Windows does what I've nicknamed core smear where a single program thread doesn't sit still, and tends to occupy partially both threads of a core. If there is much of that going on, with multiple software, it can become difficult to tell what is where. 

 

On that note, it might also be interesting if it could offer a hybrid view of top processes using CPU time (minor ones are unimportant), and also colour code them so you can see on which core(s) it is running, potentially allowing manual optimisations via affinity. This is beyond Task Manager. Resource Monitor is closer but not quite there.

The overhead that might cause to allow for it might not be worth it, I could see it working for processes that have manual affinity applied but per process tracking like that for everything might just chew up resources, dunno might be a lot simpler than I'm thinking.

[Sniperfox47]

35 minutes ago, porina said:

[...] tends to occupy partially both threads of a core. If there is much of that going on, with multiple software, it can become difficult to tell what is where. 

Yeah because a "hardware thread" or "logical core" isn't really a physical thing and a "thread" doesn't have utilization, a core does.

 

If your primary thread on a core has perfect optimal utilization with no waiting time, your "other thread" is also 100% utilized even if it's empty.

 

I honestly wish people would drop all the talk about logical cores since it just serves to confuse people. Yes SMT is great, but representing it as additional cores is super inaccurate.

[leadeater]

47 minutes ago, Sniperfox47 said:

Yeah because a "hardware thread" or "logical core" isn't really a physical thing and a "thread" doesn't have utilization, a core does.

 

If your primary thread on a core has perfect optimal utilization with no waiting time, your "other thread" is also 100% utilized even if it's empty.

 

I honestly wish people would drop all the talk about logical cores since it just serves to confuse people. Yes SMT is great, but representing it as additional cores is super inaccurate.

I think that comment was more in reference to an application process having multiple thread handles being executed on two logical threads of the same physical core. Though I'm not sure how that makes it harder to tell what processor threads are being used for an application, fairly easy if there is only a single application with high utilization but once you have 2 or more knowing what is where is near impossible unless there is known behavior that you can look for.

[AlTech]

Well at least it means people with high core count CPUs don't have slow netflix .

[Dabombinable]

15 hours ago, leadeater said:

CPU utilization has never been a metric for actual CPU performance utilization, it's a time slice usage metric as in how often is it doing something, anything. An application thread could be using a CPU thread time wise 100% but be doing actually nothing, hence why HT/SMT was such a nice thing when it got introduced and applications back then rather often were spin waiting for resources.

Pentium M with HT would have been godlike, with the difference it made to Northwood (Pentium M770/2.13 GHz  trades blows with a P4 HT/DT 3.2GHz and 631 3GHz, without HT)

[Septimus]

On 5/2/2018 at 10:24 AM, Crunchy Dragon said:

I thought it was common knowledge that Task Manager wasn't 100% accurate, which is why people always recommend something else like CPU-Z.

How much of a margin of error are we talking about here? 2-3% or something like 5-10%? And how much better is CPU-Z?

[SpaceGhostC2C]

1 hour ago, Septimus said:

And how much better is CPU-Z?

CPU-Z doesn't display CPU utilization.

[Bananasplit_00]

Any time the CPU can't be used for other stuff, I'd say is it being used even if its just waiting, I still can't use it for other stuff so why report it as not used?