Are "CPU Load" and "CPU Utilization" the same thing?

[HungryHamster]

This might be a dumb question, but are "CPU Load" and "CPU Utilization" the same thing? It appears they are different because the graph in Corsair iCUE for CPU Load is usually around 1-5% but when I check CPU Utilization in the task manager it's much higher.

[freeagent]

I would say yes.. 

[rcmaehl]

6 minutes ago, HungryHamster said:

This might be a dumb question, but are "CPU Load" and "CPU Utilization" the same thing? It appears they are different because the graph in Corsair iCUE for CPU Load is usually around 1-5% but when I check CPU Utilization in the task manager it's much higher.

Technically, yes. Implementation wise, it depends on how each is being calculated.

 

There are a few different ways they could be different but still "right":

• Kernel + User CPU usage vs User only CPU usage
• Core 0 usage vs Overall usage
• Sampling frequency + average calculations

How far of a difference are we talking? Is there any additional information?

[Arethusa]

In what units are each measured in? Assuming you have time on the x axis and the variable of interest is on the y axis. If its just percentage with no other information, then it's not much to go off of. 

[YoungBlade]

Technically, they are different, but in typical enthusiast circles, the two are used interchangeably. I'm sure some people somewhere are upset about this fact, but if you hear "load" and "utilization" or "usage," the person probably just means "the number the CPU or core is reporting in Task Manager/HWInfo/RTSS."

[HungryHamster]

28 minutes ago, rcmaehl said:

Technically, yes. Implementation wise, it depends on how each is being calculated.

 

There are a few different ways they could be different but still "right":

• Kernel + User CPU usage vs User only CPU usage
• Core 0 usage vs Overall usage
• Sampling frequency + average calculations

How far of a difference are we talking? Is there any additional information?

Thanks for your detailed response. For example, the CPU Load while playing a game is usually 1-3% but I have seen it jump momentarily to 15-18% when first starting a game. In contrast, the CPU usage is around 40-50%. When just doing stuff on the internet the load is constantly at 1% and the CPU utilization is 5-9%. It still seems strange to me.

[unclewebb]

@HungryHamster

 

On Intel CPUs, CPU Load and what the Task Manager reports for CPU Utilization are not even close to the same thing. Everyone just assumes that they are the same but they are not. 

 

For Intel, CPU Utilization is calculated based on the base frequency of the processor. Most Intel CPUs in recent memory use Turbo Boost when they are running at full speed. The higher the percentage of turbo boost a CPU uses, the higher the CPU Utilization will be reported compared to actual CPU Usage. 

 

Here is a good example. The TS Bench test is used to load 10 Threads of a 20 Thread CPU. CPU Usage is 50% plus any Windows background tasks. ThrottleStop shows this as the CPU spending 50.2% of its time in the C0 state working on this task. The Task Manager shows this as a consistent 69% Utilization.

 

 

Here is why.

 

The 10850K has a base multiplier of 36. When this is increased to 50, the Utilization data is also scaled. 

 

50.2% Usage X ( 50 / 36 ) = 69.7% Utilization

 

Everyone likes looking at those Windows graphs to judge how much CPU usage is happening. Only the Task Manager Details tab shows CPU usage. The nice Task Manager graphs are graphing meaningless information on computers with Intel turbo boosted CPUs. 

[HungryHamster]

34 minutes ago, Arethusa said:

In what units are each measured in? Assuming you have time on the x axis and the variable of interest is on the y axis. If its just percentage with no other information, then it's not much to go off of. 

Yes, unfortunately it's just percentage.

[rcmaehl]

14 minutes ago, unclewebb said:

@HungryHamster

 

On Intel CPUs, CPU Load and what the Task Manager reports for CPU Utilization are not even close to the same thing. Everyone just assumes that they are the same but they are not. 

 

For Intel, CPU Utilization is calculated based on the base frequency of the processor. Most Intel CPUs in recent memory use Turbo Boost when they are running at full speed. The higher the percentage of turbo boost a CPU uses, the higher the CPU Utilization will be reported compared to actual CPU Usage. 

 

Here is a good example. The TS Bench test is used to load 10 Threads of a 20 Thread CPU. CPU Usage is 50% plus any Windows background tasks. ThrottleStop shows this as the CPU spending 50.2% of its time in the C0 state working on this task. The Task Manager shows this as a consistent 69% Utilization.

 

 

Here is why.

 

The 10850K has a base multiplier of 36. When this is increased to 50, the Utilization data is also scaled. 

 

50.2% Usage X ( 50 / 36 ) = 69.7% Utilization

 

Everyone likes looking at those Windows graphs to judge how much CPU usage is happening. Only the Task Manager Details tab shows CPU usage. The nice Task Manager graphs are graphing meaningless information on computers with Intel turbo boosted CPUs. 

This is also very true. 

[porina]

22 minutes ago, unclewebb said:

For Intel, CPU Utilization is calculated based on the base frequency of the processor. Most Intel CPUs in recent memory use Turbo Boost when they are running at full speed. The higher the percentage of turbo boost a CPU uses, the higher the CPU Utilization will be reported compared to actual CPU Usage. 

 

[snip]

 

The 10850K has a base multiplier of 36. When this is increased to 50, the Utilization data is also scaled. 

 

50.2% Usage X ( 50 / 36 ) = 69.7% Utilization

 

Everyone likes looking at those Windows graphs to judge how much CPU usage is happening. Only the Task Manager Details tab shows CPU usage. The nice Task Manager graphs are graphing meaningless information on computers with Intel turbo boosted CPUs. 

I'm struggling to follow this although I can replicate what you describe. I had taken Windows reporting to be CPU time, but doesn't consider throughput. e.g. non-linear effect when all cores are effectively loaded and you are tapping into what's left from HT.

 

In my 7920X system, if I run Prime95 50% threads it drops to base clock due to AVX-512. I get about 50% usage shown in Windows as expected. If I repeat that with AVX disabled, the CPU boosts around 3.8 GHz, compared to 2.9 base. By your calculation method that's around 66%, and I see around 67% reported in task manager. Close enough.

 

Do you have a reference to another description of this effect where I can read up more on it? Also was there a change in Windows at some point, because this behaviour only started some time during the life of Win10. Early Win10, use 50% threads, see 50% usage regardless of clock. So MS changed something, sometime. If enabled to be shown the excess may be related to "kernel time" but I never proved this.

 

The big problem I have with this measurement is that I think it would be valid if and only if CPU usage could go above 100%. As it stands I can hit 100% on Task Manager when the CPU isn't IMO fully loaded. I haven't checked I can actually get the expected throughput as otherwise that'll be a big performance loss.

 

Edit: to answer myself, I found the following which kinda describes it too. https://aaron-margosis.medium.com/task-managers-cpu-numbers-are-all-but-meaningless-2d165b421e43 So it seems Windows did use %time in past although it seems ambiguous when the switch to "utilisation" was made, might have been before Win10. So it seems best way to read the Windows reported value is it is some number bigger than what you intuitively think.

[unclewebb]

@porina

 

Here is another Windows reporting issue that no one ever seems to question.

 

 

How is it possible to have the CPU running at more than 100% of its Maximum Frequency especially when the CPU is idle? This is only possible if the base frequency is what Microsoft is using to calculate the Maximum Frequency. My old 4th Gen laptop with a 4700MQ can use the 36 multiplier when 1 core is active. When running at its base frequency, it uses the 24 multiplier. 

 

36 / 24 = 150%

 

I can show you examples of this misleading data but I have no idea why Microsoft decided to compare everything to the base frequency and not to the maximum turbo boost frequency, Intel CPUs mostly run at the max turbo frequency.

 

When Microsoft first started doing this, desktop CPUs only used a small percentage of turbo boost. No one really noticed or cared that CPU Utilization and CPU Usage were not in sync. Some Intel mobile CPUs have very low base speeds and very high turbo speeds. The inflated looking Utilization data becomes a lot more obvious then. 

 

On my above example with 10 of 20 threads loaded, when I select Disable Turbo in ThrottleStop and run the CPU at its base frequency, CPU Usage and CPU Utilization are exactly the same. If I use ThrottleStop and slow the CPU down to a crawl while the TS Bench is still running, the CPU Usage is still exactly the same at 50% but the Utilization number goes way down. When forcing the CPU to run the TS Bench at 800 MHz the Utilization is,

 

(800 MHz / 3600 MHz) X 50% Usage = 11% Utilization 

 

On my desktop computer with a 50 multiplier and a 36 base multiplier, when CPU Usage is 72% or higher, Windows scales this up and reports 100% Utilization. The problem then is that 72% Usage or 80% or 90% or 100% CPU Usage will all be reported exactly the same as 100% Utilization. The Task Manager just caps the maximum Utilization number at 100%. This leaves some users thinking that their CPU is more loaded than it actually is. This data has probably encouraged more than one user that it is time for a faster CPU with more cores to handle this imaginary load. 

 

5 hours ago, porina said:

Do you have a reference to another description of this effect

I am not sure what you are looking for. This issue has been around for years. No one has ever been interested enough to stop and ask Microsoft why they decided to display and graph utilization data in the Task Manager instead of CPU usage. The Task Manager Details tab shows data for CPU Usage. 

[porina]

6 hours ago, unclewebb said:

I can show you examples of this misleading data but I have no idea why Microsoft decided to compare everything to the base frequency and not to the maximum turbo boost frequency, Intel CPUs mostly run at the max turbo frequency.

 

When Microsoft first started doing this, desktop CPUs only used a small percentage of turbo boost. No one really noticed or cared that CPU Utilization and CPU Usage were not in sync. Some Intel mobile CPUs have very low base speeds and very high turbo speeds. The inflated looking Utilization data becomes a lot more obvious then. 

I guess I don't use Resource Monitor enough to ever notice that. I'm not sure it is true that Intel CPUs run at their max turbo most of the time for many generations now. Maybe not since we left the quad core era? Take the 6700k, that was 4.0 GHz base, 4.2 GHz turbo but many mobos would overclock it for you so all core also ran 4.2, which was not a big jump. Even in Coffee Lake era, that gap was much bigger. 8086k was 5.0 GHz turbo, but all core was 4.3 GHz. Then you get even bigger differences in mobile CPUs.

 

May MS wanted to avoid users going "why isn't my CPU going to 100%?". Hitting turbo clocks is not guaranteed. 

 

6 hours ago, unclewebb said:

I am not sure what you are looking for. This issue has been around for years.

The link I edited in did go to more detail which is what I was after. It was still ambiguous when it started, that link suggest it might even been as early as Win 8.x. I don't recall exactly when I noticed it, other than it was in Win 10 era. From a user perspective, it seems like it is just broken display so work done is still fine.

[unclewebb]

8 hours ago, porina said:

6700k

You are correct. The default turbo speed of the 6700K was only 5% faster than the base frequency. If you started overclocking by increasing the maximum turbo speed, this percentage would increase and the difference between CPU utilization and CPU usage data would also increase. The 6700T with its 35W TDP used a lower base frequency so at max turbo boost the difference was (3.60 GHz / 2.80 GHz) or 28.6%. 

 

The mobile CPUs tend to use a lower base frequency. This reporting problem is very noticeable with my 4700MQ at (3.60 GHz / 2.40 GHz) or 50.0%. 

 

Modern low power mobile CPUs like the 10710U use a very low base frequency and a very high maximum turbo frequency. 

 

https://ark.intel.com/content/www/us/en/ark/products/196448/intel-core-i710710u-processor-12m-cache-up-to-4-70-ghz.html

 

4.70 GHz / 1.10 GHz is a difference of 327%. That kind of difference is going to grossly inflate the Utilization data so it is completely unusable.