What’s the definition of thermal throttling?

[Spectrum]

It would need some modification or people would keep asking why it keeps being throttled down to 3.2GHz when it's at low usage.

Yeah - I wasn't being completely serious. I'm not that much of a conspiracy theorist.
But regarding your point, most chips clock down way below the rated "base" frequency when idle. So I'm not sure what you say would actually be apparent.
The 'base' speed of my current CPU is 2Ghz, but it is sitting at 0.8-1.4 as I type this.

 

[pl1984]

The Intel marketing term "Turbo boost" was coined many years ago to hide the fact that their chips couldn't actually run at the frequency everyone wanted- they were throttled down almost immediately to keep the CPU from melting. But advertising a "Turbo" frequency that was much higher sure looked good on paper - even if it only could run at the frequency for a very short while.

This is incorrect. When processors were transitioning from single to multiple cores consumers were faced with a delimma:

1. Purchase a single core CPU with a higher clock speed
2. Purchase a multiple core CPU with lower, per core clock speeds

This was because a given CPU is limited by its power envelope. Higher clock speeds consume more power. Likewise more cores mean higher power consumption. You can't increase both clock speed and core count without increasing the power envelope. Therefore one had to decide if single or multiple core performance was better for their workload and buy accordingly.

In an effort to eliminate the need to make such a choice Intel (I'll use them for the sake of this discussion) decided to make processors that dynamically adjusted the clock frequency to remain within the defined power envelope. When only a single core was in use the processor could increase the clock frequency up to the point where the power envelope was reached. When multiple cores were in use the processor would clock those cores appropriately to maintain the power envelope. Thus the processor would dynamically adjust the clock speed based on the thread workload. Less threads typically meant higher clock speed (up to the maximum boost speed), more threads typically means lower clock speeds (down to the base speed).

Turbo boost is exactly what its name state: A boost to the clock speed as the power envelope permits.

 

[Stephen.R]

But regarding your point, most chips clock down way below the rated "base" frequency when idle. So I'm not sure what you say would actually be apparent.
The base speed of my current CPU is 2Ghz, but it is sitting at 0.8-1.4 as I type this.

Interesting. Is that a laptop or a desktop?

My understanding is that this would be "SpeedStep" in action, I didn't know it's even available on desktop CPUs until today.

 

[pl1984]

Interesting. Is that a laptop or a desktop?

My understanding is that this would be "SpeedStep" in action, I didn't know it's even available on desktop CPUs until today.

It is not uncommon for a CPU to clock down below its base frequency when it is idle or lightly loaded. For example my z440 I'm typing this response on is currently showing 1.1GHz clock speed.

 

[Spectrum]

Interesting. Is that a laptop or a desktop?

My understanding is that this would be "SpeedStep" in action, I didn't know it's even available on desktop CPUs until today.

It is the 2011 mini - so yes, it is a "laptop" CPU. I didn't realise this behaviour was unique to laptop parts.
Is the processor in the 2016 4K iMac also "laptop"? i7-5775R. I am pretty sure this also drops down way below its base speed of 3.3 Ghz when idle.

 

[LeeW]

My understanding is that this would be "SpeedStep" in action, I didn't know it's even available on desktop CPUs until today.

Not looked at the spec sheet to see if speed step is there but it's going to clock down at idle. See screenshot, this is my new MM with the i7. Jumping up at the end only as I took the screenshot. It's about where I would expect at idle.

Edit: Yes, it has Enhanced Intel Speed Step (EIST) which intentionally throttles it back as low as 800GHZ when the system is not busy.

 

[Stephen.R]

I didn't realise this behaviour was unique to laptop parts.

Well, it isn't. I thought it was, but I was incorrect.

Is the processor in the 2016 4K iMac also "laptop"

I believe all recent (non Pro) iMacs use 'mobile' Intel processors?

Not looked at the spec sheet to see if speed step is there but it's going to clock down at idle

Fair enough.

 

[Spectrum]

This is incorrect. When processors were transitioning from single to multiple cores consumers were faced with a delimma:

1. Purchase a single core CPU with a higher clock speed
2. Purchase a multiple core CPU with lower, per core clock speeds

This was because a given CPU is limited by its power envelope. Higher clock speeds consume more power. Likewise more cores mean higher power consumption. You can't increase both clock speed and core count without increasing the power envelope. Therefore one had to decide if single or multiple core performance was better for their workload and buy accordingly.

In an effort to eliminate the need to make such a choice Intel (I'll use them for the sake of this discussion) decided to make processors that dynamically adjusted the clock frequency to remain within the defined power envelope. When only a single core was in use the processor could increase the clock frequency up to the point where the power envelope was reached. When multiple cores were in use the processor would clock those cores appropriately to maintain the power envelope. Thus the processor would dynamically adjust the clock speed based on the thread workload. Less threads typically meant higher clock speed (up to the maximum boost speed), more threads typically means lower clock speeds (down to the base speed).

Turbo boost is exactly what its name state: A boost to the clock speed as the power envelope permits.

Yes..But Intel also specify all chips as having some degree of boost (often significant) also with a multi-threaded workload.
For example, the i7-8700B is base of 3.2 and boosts of 4.6, 4.5, 4.4, 4.3, 4.3, 4.3 for progressive core usage.
Those are the specs. So, naively, why would one not expect to get that level of performance?

If the specs were: 4.6, 4.0, 3.6, 3.2, 3.2, 3.2 (which may be more achievable for a mid-range cooling system) then there would be no discussion.

The point is the spec sheet says 4.3 for 6-core, not 3.2 for 6-core.

Perhaps it would be better if the spec said: 6-core max turbo 4.3 @ 130Watts, so that it is clear that if you want to get the absolute max performance out of the CPU (all core boost) you need a cooler that is specced 2x above the rated TDP. And without said cooler, the chip is, essentially, being throttled to the tune of 30%.

 

[Stephen.R]

specify all chips as having some degree of boost

They specify them as having some degree of boost potential.

If it were guaranteed, it would be the base frequency.

 

[LeeW]

just for comparison, one shows it responding as I do basic things, email, browsing. The other is a cinebench test. As you can see once it ramps up and the fan kicks in it settles comfortably above base clock around 3.6 GHZ.

In any event it really all depends on what performance you need for the things you do.

 

[Spectrum]

just for comparison, one shows it responding as I do basic things, email, browsing. The other is a cinebench test. As you can see once it ramps up and the fan kicks in it settles comfortably above base clock around 3.6 GHZ.

In any event it really all depends on what performance you need for the things you do.

Very helpful and interesting graphs LeeW. They tell me a few of things:

1/ The temperature increase on these chips is dramatic - reaching 100C within the first few seconds of maximum power draw.
2/ The max all core turbo is about 4.3 Ghz (as specified by Intel) and requires about 110 Watts of power.
3/ The 2018 mini cooling system is good for about 75 Watts of cooling
4/ 75 watts are required for an all-core turbo of 3.6Ghz
5/ A bigger, costlier cooling system specced for 110 Watts would likely enable an all core turbo of 4.3Ghz indefinitely.

 

[LeeW]

The temperature increase on these chips is dramatic - reaching 100C within the first few seconds of maximum power draw.

I have said in other posts that the fan curve needs to be adjusted, the new cooling is certainly much better but then it needs to be given the significant jump in processor. What I have noticed is that when the fans ramp up it cools it perfectly well, Apple have chosen the acceptable temp to be in the low 90's, that is where it keeps it under load but the fan has much more to give.

Much like the the 2018 MBP it waited until the temp was at 100C before the fans ramped up, it's much better now so perhaps it will be tweaked soon.

 

[pl1984]

just for comparison, one shows it responding as I do basic things, email, browsing. The other is a cinebench test. As you can see once it ramps up and the fan kicks in it settles comfortably above base clock around 3.6 GHZ.

In any event it really all depends on what performance you need for the things you do.

As Stephen.R stated: It's a potential, not a guarantee. The fact a processor can run above its base frequency with all cores active does not change anything.

 

[Spectrum]

I really don't know what we are arguing about. The graphs clearly show that full max core boosts make it hit >110 Watts. And that the current cooling system is good only to reach equilibrium at 75 Watts. Which is the heat generated by 6-cores @ 3.6 Ghz.

The fact that this is above the base clock of 3.2 Ghz isn't relevant to anything. The CPU design itself is actually limited to 6-cores @ 4.3 Ghz.

The 'base' clock speed looks to be just defined by the maximum 6-core speed that hits the stated TDP of 65 Watts (or lower).

i.e. The i7-8700B processor has a 6-core speed of 6x3.2 Ghz @ 65 Watts (or lower).
The tests show it also has a 6-core speed of 6x3.6 Ghz @ 75 Watts (the apparent stable limit of the cooling system)
And 6-core speed of 4.3Ghz @ 120 Watts (well above the stable limit of the cooling system).

The same processor could have been marketed as a 35 W processor with base clock of 2.3 Ghz. Obviously with a cooling system fit for only 35 Watts, the max all-core turbo is now going to be ~2.3 Ghz. And no one would say it is throttling...Even though the CPU would be identical...and it is clearly being throttled relative to the performance it could get with better cooling.
[doublepost=1542398185][/doublepost]

I really don't know what we are arguing about. The graphs clearly show that full max core boosts make it hit >110 Watts. And that the current cooling system is good only to reach equilibrium at 75 Watts. Which is the heat generated by 6-cores @ 3.6 Ghz.

The fact that this is above the base clock of 3.2 Ghz isn't relevant to anything. The CPU design itself is actually limited to 6-cores @ 4.3 Ghz.

The 'base' clock speed looks to be just defined by the maximum 6-core speed that hits the stated TDP of 65 Watts (or lower).

i.e. The i7-8700B processor has a 6-core speed of 6x3.2 Ghz @ 65 Watts (or lower).
The tests show it also has a 6-core speed of 6x3.6 Ghz @ 75 Watts (the apparent stable limit of the cooling system)
And 6-core speed of 4.3Ghz @ 120 Watts (well above the stable limit of the cooling system).

The same processor could have been marketed as a 35 W processor with base clock of 2.3 Ghz. Obviously with a cooling system fit for only 35 Watts, the max all-core turbo is now going to be ~2.3 Ghz. And no one would say it is throttling...Even though the CPU would be identical...and it is clearly being throttled relative to the performance it could get with better cooling.

Take the new MBAir processor - 1.6 GHz base and 3.6 Ghz turbo. A 7 Watt chip.
I expect that 7 watts is probably the power draw of the dual-core chip at 1.6 Ghz base speed.
There is no way it is using 7 Watts at 3.6Ghz.
Therefore, to get faster speeds for any length of time, you needs a cooling system much better than 7 Watts.
But let's say the sustained dual core speed in the MBAir is only 2 Ghz...would we not conclude that it is significantly throttled?
[doublepost=1542398245][/doublepost]Anyway...this is getting tiresome...

 

[droog]

So much gas lighting in here.

Chips can't run nearly as well as other similarly priced and spec'd machines with proper cooling. Apple is willing to sacrifice performance for aesthetics. This is already known. End of story.

 

[LeeW]

So much gas lighting in here.

Chips can't run nearly as well as other similarly priced and spec'd machines with proper cooling. Apple is willing to sacrifice performance for aesthetics. This is already known. End of story.

The vast majority of customers, who don't care about these discussions simply get on and use their devices and are happy that they can achieve what they want on them. That will represent 80% or more.

For the remaining 20% there are these discussions, it's not about gas lighting, we all know the CPU's will have performance constraints due to the form factor and Apple's interpretation of things. Many simply want to know what those constraints are before purchasing and whether it fits their needs before making a significant financial commitment.

So it's fine for you to come along with your "This is already known. End of story.". Now that you have done that, you can chill and leave others to continue the discussion as they want.

 

[Flynnstone]

In the “old” days processor margin was simpler. A processor was rated and limited to a particular core frequency. In reality, they could run faster (& hotter). At higher ambient temperatures this margin would be reduced. The over clockers came into being using this margin to their advantage. Bigger cooler heat sink to keep die temperature down.
Then someone said “hey we’ll be doing 20GHz in a couple years”. As we all know that hasn’t happened. But obviously most like higher clock rates. More is better!
About this time, AMD (I think) created the first thermal throttling. Should still be some videos of people popping heatsinks off the processors under load. Intels would go poof and AMDs would thermal throttle. Processor safety feature.
Later someone got the good idea to use this to advantage. Most of the time, most people’s CPU’s are running under 10%. Everyone check you cpu usage! I think most will be under 10%. Perhaps most are at 1%!
So run the processor at a reduced clock rate to be power efficient. Since the die is cooler and it has some thermal mass, we can boost the clock speed briefly to accomplish a small task. A slight increase in performance. And a marketing feature!

 

[pl1984]

The 'base' clock speed looks to be just defined by the maximum 6-core speed that hits the stated TDP of 65 Watts (or lower).

You don't have to guess as to what it is as Intel defines it. From the Core i7-8700B ARK page:

"Processor Base Frequency describes the rate at which the processor's transistors open and close. The processor base frequency is the operating point where TDP is defined. Frequency is measured in gigahertz (GHz), or billion cycles per second."

https://ark.intel.com/products/134905/Intel-Core-i7-8700B-Processor-12M-Cache-up-to-4-60-GHz-

Turbo Frequency:

"Max turbo frequency is the maximum single core frequency at which the processor is capable of operating using Intel® Turbo Boost Technology and, if present, Intel® Thermal Velocity Boost. Frequency is measured in gigahertz (GHz), or billion cycles per second."

 

[droog]

The vast majority of customers, who don't care about these discussions simply get on and use their devices and are happy that they can achieve what they want on them. That will represent 80% or more.

For the remaining 20% there are these discussions, it's not about gas lighting, we all know the CPU's will have performance constraints due to the form factor and Apple's interpretation of things. Many simply want to know what those constraints are before purchasing and whether it fits their needs before making a significant financial commitment.

So it's fine for you to come along with your "This is already known. End of story.". Now that you have done that, you can chill and leave others to continue the discussion as they want.

Every chip has a unique tdp and you can either design with that in mind or you can "Apple it" and not care at all. If you are concerned about maximizing your chips capabilities, don't buy apple hardware.

 

[pl1984]

The vast majority of customers, who don't care about these discussions simply get on and use their devices and are happy that they can achieve what they want on them. That will represent 80% or more.

I disagree. While the vast majority of customers may not care about the technical details they will care if their higher cost Core i7 system doesn't perform as fast as a Core i5 system.

 

[LeeW]

If you are concerned about maximizing your chips capabilities, don't buy apple hardware.

Indeed and that is where some people need to get to, I would always hope that someone who knows and understands why they must be able to maximise their chip performance will also understand why something like a MM is not going to help them achieve that goal. For everyone else they need to understand that thermal throttling is not something that needs to keep them awake at night.

I disagree. While the vast majority of customers may not care about the technical details they will care if their higher cost Core i7 system doesn't perform as fast as a Core i5 system.

Appreciate opinions vary on that but my experience of general users tells me that they largely measure performance based on how much quicker the next system is than their last, they buy based on the upper element of their budget and most often end up with a system that is overpowered for their needs anyway.

 

[pl1984]

Appreciate opinions vary on that but my experience of general users tells me that they largely measure performance based on how much quicker the next system is than their last, they buy based on the upper element of their budget and most often end up with a system that is overpowered for their needs anyway.

None of which is in opposition to what I said.

 

[brentsg]

The advertised clock speed can be whatever Intel wants it to be, up to its maximum capability before it cooks itself to death.

The Intel marketing term "Turbo boost" was coined many years ago to hide the fact that their chips couldn't actually run at the frequency everyone wanted- they were throttled down almost immediately to keep the CPU from melting. But advertising a "Turbo" frequency that was much higher sure looked good on paper - even if it only could run at the frequency for a very short while.

So it looked good on paper. "Throttling" and slowing was replaced with the term "Turbo Boost" to trick the end user into thinking they were getting something extra instead of taken away.

It's all marketing speak. The CPUs can run at the speed they can run at. It's what you WANT them to run at that's causing all the commotion. And that's intel's fault.

Except there is the baseline speed for everyone to see. Whether you like it or not, turbo has practical benefits for everyone. It is much more than marketing.

I agree, the fact is the Mac Mini and most of Apple's product line doesn't have good enough cooling to sustain Turbo. Is it unrealistic to expect it especially for the pro market where this new Mini is now appealing to? No, in the enthusiast market for example it's pretty much standard practice to have a huge heatsink or water-cooled loop to be able to get the best performance possible.

Apple just prioritizes form over function, thinness, looks take priority over cooling/throttling/noise.

This is a good read:
https://www.notebookcheck.net/Opinion-It-s-time-we-talked-about-throttling-in-reviews.234232.0.html

There’s a reason that people get base pay and bonus pay. You can count on the base pay, but if certain conditions aren’t met, that bonus money isn’t happening. It’s the same thing here.

 

[Spectrum]

My feeling is that Intel should specify CPUs with their associated wattage at: idle, base speed, base speed all core, max turbo single core, max turbo all core.

And ideally, Apple or whoever is selling the system should rate the maximum cooling capable, or the max sustained single core and all core CPU speed it can handle.

They won't, of course, because it might look bad...but at least one could accurately compare systems.

 

[whichimac]

1. does any of this have any relevance to how 'thermals' affects the speed of mini?

it appears? that mini is fine?

dated 18.11

- https://www.reddit.com/r/macmini/comments/a0pji7/2018_mac_mini_cinebench_r15_info_with_egpu_and/

- https://www.reddit.com/r/macmini/comments/9x7k57/mac_mini_thermal_throttle_test_i7/

- https://appleinsider.com/articles/1...tling-and-performance-in-the-2018-i7-mac-mini

- https://www.reddit.com/r/apple/comments/9vdq7c/mac_mini_i7_ive_tested_thermal_performance/


2. what is the most conclusive test currently out there on how much 'thermals' affects the speed of mini?