0

I noticed a discrepancy in Task Manager between the performance tab and the details tab. Performance tab shows 14% utilization, while the numbers in the details tab add up to about 5%. I know there are small rounding errors in the details tab, but this question is not about that. One process osd.exe shows 4% in the details tab, while it shows 10.2% in Process Explorer. This screenshot shows both open simultaneously. Also at the same time, the Task Manager processes tab shows 9.2% for this process.

The total utilization shown at the bottom of Process Explorer matches the performance tab of Task Manager.

Also, Python's psutil.cpu_percent() shows 6.6% which is pretty close to the sum of the Task Manager details tab since the details tab rounds to the nearest integer.

Apparently these different views are using 3 different methods to calculate CPU consumption. I would love to find some documentation showing the differences. This is on Windows 10.

I also checked it with high CPU utilization by running Prime95 torture test:
Task Manager details tab 90%
Task Manager processes tab 83%
Python psutil 100%
Process Explorer 83%

I believe Python and Task Manager details tab are accurate because Prime95 is supposed to push the CPU to its limit, but I'm not sure.

enter image description here

Elliott B
  • 1,347
  • 5
  • 17
  • 44

1 Answers1

0

Addressing Elliot's initial concern, "I noticed a discrepancy in Task Manager between the performance tab and the details tab.", I've had the same concern for years. And in experiencing it again today, even still on Windows 11 24h2, I dug around and found first this thread and other places where people were asking--but without any clear answer, until I found this offered explanation:

Task Manager’s Processes and Performance tabs now use the “% Processor Utility” counter as the basis for their CPU numbers, rather than the “% Processor Time” counter Task Manager had relied upon and that is still used by Task Manager’s Details tab and by Sysinternals Process Explorer.

It's from a helpful article which addresses this matter (with more detail) by Aaron Margosis (the italics above are mine). Note how he also tied it into ProcExp, which was another part of Elliot's question. Time and closer assessment by others will confirm the accuracy, but at least for now it seems a reasonable explanation.

charlie arehart
  • 292