Slow and irregular sensor update while partial CPU loads

[Furda]

First, I want to thank you for providing HWiNFO, absolutely great tool, thank you very very much!


TLDR;
Sensors do not update regularly as defined by the interval setting, sensors update seems stuck/blocked when CPU is under partial heavy load.

I was doing some tests with different CPU loads with the "Stress CPU" function of CPU-Z (please see attached screenshot for an example).

"Threads" option (combobox) allows 1,2,4,8,12,14,16,18, or 20 threads with my 12700k.

Important:
The issue occurs only with 2,4,8,12,14,16,18 threads.
It does *not* occur with 1 or 20 threads.

I use a sensor update interval of 1000ms and noticed that sensors took more time to update than normally. The counter on the very bottom right of the sensors window was not updating (almost) every second as usual, it took several seconds between each update. It seems it updates about every 3 seconds on an average, sometimes faster, sometimes (a lot) slower.

A log to .csv showed the same, the entry interval ranges from 1 to 6 seconds (mainly). Please see the "hwinfo_log.csv" as an example (recorded with debug mode off).
At 2mins 41s in this log, it seems to "recover" from the issue and writes a log every second as expected.
(Another example of this "recovery" follows below)

I've attached a second log (please see the file "log_debug_on.csv"), it was recorded during debug mode alongside with "HWiNFO64.DBG".
Time about 0 to 30 seconds of the log were made with stress CPU with 1 thread that took 10 seconds.
Time about 30 to 60 seconds of the log were made with stress CPU with 20 threads that took 10 seconds.
Time about 60 to 90 seconds of the log were made with stress CPU with 12 threads that took 10 seconds.
(You'll see the change in CPU usage % between idle and the three stress periods.)

Windows was responsive during every stress test, also with full 20T on stress.
Other tools showing/monitoring some system parameters running at the same time do update without issues (where HWiNFO seems locked).

Important:
Then, after a few minutes, HWiNFO seems to somehow recover itself and then the sensors are updated every second as expected.
I've attached an example for that, "log_recovers_itself.csv", first it shows the issue with slow update, then after about 1min 10s, it recovers, updates every second as expected.
I had another case, where the "recovery" time took about 2min30s.

I did some tests with older versions of HWiNFO I had around too, v7.24, v7.30, v7.40, v7.42, v7.46 seem to have the same issue as 7.47 beta 5120 I'm using now. I deleted the HKCU\SW\HWiNFO64-key in registry between every test.

For a test, I disabled "CorsairLink" support (as I am using two Corsair devices), but the issue persisted.

I've "Highlight changing values" usually off, for a test I enabled it, no change.

Well so far, I assume that this is not a "feature" of HWiNFO, as the issue seems to "recover" itself after sometime. Furthermore, I don't think its a Windows issue, as the system remains very well resposive all the time, and other system info tools update values where HWiNFO seems to be stuck. I've run HWiNFO without any other software to check it's not something else interfering.

If you need any other infos, please let me know.
Thank you very much for your kind assistance.

Attachments:
- CPU-Z_stress.png: FYI
- hwinfo_log.csv: log with debug mode off, CPU load almost 100% for about 3 mins
- log_debug_on.csv: log with debug mode on, three stress tests with 1T, 20T, 12T for each 10 seconds.
- HWiNFO64.dbg: debug log
- log_recovers_itself.csv:
- HWiNFO64.ini: my actual settings, debug mode off (can provide .reg if you need it)

 

[Martin]

I don't see any excessive delays in HWiNFO itself, so I think this might be due to how the HWiNFO is scheduled by Windows along with CPU-Z.
Try to increase the priority of HWiNFO64 in Task Manager to see if that will help.

 

[Furda]

Thank you for your instant response.

To be honest, I though about that, but I never ever thought that Windows 11 would schedule that worse, with a 12th Gen Core CPU, with Intel's "Thread Director". Escpecially with 2T only, on a 12C/20T CPU. Thus, I did not test that. Up to now.

As soon as I open the task manager, "magically" the sensors update a lot more frequent, almost every second. And yes, if I decrease prio, updating gets even slower. If I increase prio, it's better.

Well, shame on me I didn't quick test that, although I should have knew it, wasting your time sorry (and mine). Shame on me trusting Microsoft and Intel for a proper scheduling, in the year 2023.
(It's highly disappointing, and nothing else than a huge shame how bad scheduling works!)

Martin, do you have an idea, why everything else seems well responsive, but HWiNFO isn't, at least in my case?

Any others seeing such an effect? Couldn't really believe I'm the first reporting that, should have been noticed by almost everybody doing CPU load tests isn't it?

 

[Martin]

This is not an issue with "Thread Director" but scheduling of tasks in Windows. If the main HWiNFO thread is scheduled on the same core where the CPU-Z runs the high-load thread, then HWiNFO gets less time-slot if it running at the same or lower priority.

 

[Furda]

Whoever's fault it is, Microsoft's and/or Intel's, HWiNFO isn't for sure, the scheduling is just awful. It's a pure shame how bad this is done with a socalled "state-of-the-art" OS and CPU in 2023.

It's a 12C/20T CPU, having CPU-Z running on two theads, yes two threads in CPU-Z are enough, making HWiNFO not having enough ressources to update properly - this is a pure flat shame and nothing else than an evidence of incapacity and confession of failure!

 

[Furda]

Hello,
I'm back on Win10 and there's absolutely no delay with updating HWiNFO, doesn't matter how much threads, 1T to 20T are being stressed, flawlessly smooth, as usual.
Win11 is abismally poor.