Intel CPU "T" model differences to "S", "K" models

MathDestruction · ‎02-04-2023

Hello,

I have a question about Intel "T" series - energy efficient - CPUs.

Two differences are the lowered TDP and the lower base clock/voltage. Thats clear so far.

But few years ago when I investigated about the "T" models I read an article that said that also the "command execution order" has another optimization style.

It was explained that a "S" or "K" CPU will try to thread instructions on a first in first out policy. Like that's very good for realtime purpose and when an operation N+1 needs the result auf operation N.

It was explained that a "T" CPU will try to maximize the utilization of the CPU and may take more time to optimize the order before handling the execution in the next cycle. Like operations that are atomic - so don't require to be computed in a special order - can be pushed to the top of the command stack for better utilization. This would be worse for realtime purpose but better for high workloads in general.

To ask in another way: Is a "S" CPU set down to the specs of the equivalent "T" CPU 100,0% the same then or does the "T" CPU still have other firmware optimizations to work more efficient?

Is this correct or was the article I did read wrong? Did I missunderstand the whole thing?

Thanks for your help!

KrissyG · ‎02-05-2023

"T" series - energy efficient "

there has been tests, that proved the 'T' version to be worse.
The reason behind it is simple, for same task the 'T' version runs slower and needs more time, therefore the whole system will use more power to complete the task.

For laptops it would work out better, but if considering system power, and not CPU alone, the higher frequency and a job done quicker - is better.

Source-> A German webpage

n_scott_pearson · ‎02-05-2023

What garbage! Select the processor based upon the usage requirements. Those bimbos chose, um, unwisely.

Just saying,

...S

MathDestruction · ‎02-09-2023

Thanks for the awnsers but it still does not clarify my exact question.

I think newer T models can boost to the same max clock like S models.

So does the T model only make decissions (clockspeed, moment of execution) by the TDP budget

or

will it smarty collect intructions and then try to boost to max clock for a shorter time than a S cpu?

Related to the picture above we assume both can go 3,6Ghz.

With the same set of instructions incoming:

Will S CPU boost to 3,6Ghz with JIT-Execution for lets say 100cycles starting at timepoint 0 with a cpu load of 50%.

Will T CPU take inital time to optimize and then start boost to 3,6Ghz for lets say 80 cycles at timepoint 20 but with a higher load of 60% ?

So both would finish at the same time but the T model used less total time and energy at max speed.

KrissyG · ‎02-09-2023

@MathDestruction wrote:

Thanks for the awnsers but it still does not clarify my exact question.

I think newer T models can boost to the same max clock like S models.

So does the T model only make decissions (clockspeed, moment of execution) by the TDP budget

or

will it smarty collect intructions and then try to boost to max clock for a shorter time than a S cpu?

Related to the picture above we assume both can go 3,6Ghz.

With the same set of instructions incoming:

Will S CPU boost to 3,6Ghz with JIT-Execution for lets say 100cycles starting at timepoint 0 with a cpu load of 50%.

Will T CPU take inital time to optimize and then start boost to 3,6Ghz for lets say 80 cycles at timepoint 20 but with a higher load of 60% ?

So both would finish at the same time but the T model used less total time and energy at max speed.

That is a complicated question.
To which i can not give an answer, but rather a hint.

The screen above, shows in gray the frequency, which remains almost unchanged.
The power limitation does not really have a correlation to the actual CPU load.

It's more like the CPU can choose between adapting the frequency and/or load in order to not exceed the TDP.
And it makes sense, considering some processes may need higher frequency at which TPD will not be exceeded , versus processes where even at lower frequency the CPU hits already TDP.

At this point it is not really possible to answer such question, bcoz you would have to specifically know, what exactly the CPU will be used for.

MathDestruction · ‎03-15-2023

Hey KrissyG,

i'm a bit late with my reply because there was a maintenance downtime for forums but thanks for your informations!