What’s the definition of thermal throttling?

[jdmagoo1]

As the title suggests, what’s the definition of thermal throttling? I ask because it seems to be muddled within some if not all of the Mac mini tests.

So my understanding is, the term thermal throttling means that the cpu will adjust its speed to below base frequency in order to lower the cpu temperature to protect itself from damage caused due to overheating.

However, all the reviews I’ve read of the i7 Mac mini are saying that it also suffers from thermal throttling. But their definition is different. I haven’t seen any reviews state that the cpu will go below base frequency, yes it reduces from the stated turbo freq but doesn’t go below base (3.2ghz). The reviews I’ve watched or read say the freq generally stabilises around 3.5ghz at around 90 - 100deg c. So this actually isn’t thermal throttling but performing above what it’s rated to.

 

[rmdeluca]

https://en.wikipedia.org/wiki/Dynamic_frequency_scaling

https://en.wikipedia.org/wiki/Intel_Turbo_Boost

From the horse's mouth:

http://web.archive.org/web/20081221...ocessor/applnots/320354.pdf?iid=tech_tb+paper

 

[jdmagoo1]

https://en.wikipedia.org/wiki/Dynamic_frequency_scaling

https://en.wikipedia.org/wiki/Intel_Turbo_Boost

But then theres this:

https://www.imore.com/base-clocks-boost-clocks-and-thermal-throttling-explained

 

[M_Scout]

I guess I lean towards calling it thermal throttling if the processor ever drops below the base clock speed with reasonable ambient temperatures. What I've seen with the 2018 mac mini i7 doesn't feel like thermal throttling to me... but I understand there are various opinions on the definition

 

[Altis]

Thermal throttling is any time the clock speed has to be reduced to prevent further temperature increase.

So it could run faster (as in the CPU is perfectly able to), but it would require quicker heat dissipation. Heat results from power consumption, and the dynamic power consumption in the CPU is determined partly by the clock speed (Power = Capacitance * Voltage^2 * Frequency).

In your example, it is thermally throttled because it is clocking down due to the temperature.

 

[M_Scout]

Thermal throttling is any time the clock speed has to be reduced to prevent further temperature increase.

So it could run faster (as in the CPU is perfectly able to), but it would require quicker heat dissipation. Heat results from power consumption, and the dynamic power consumption in the CPU is determined partly by the clock speed (Power = Capacitance * Voltage^2 * Frequency).

In your example, it is thermally throttled because it is clocking down due to the temperature.

So if it ever runs below the max turbo speed (4.6 ghz) you would consider it thermal throttling? I guess i'm curious from what measurement are we using to say the clock speed has been reduced. Not trying to argue... just curious

 

[trifid]

Thermal throttling is any time the clock speed has to be reduced to prevent further temperature increase.

So it could run faster (as in the CPU is perfectly able to), but it would require quicker heat dissipation. Heat results from power consumption, and the dynamic power consumption in the CPU is determined partly by the clock speed (Power = Capacitance * Voltage^2 * Frequency).

In your example, it is thermally throttled because it is clocking down due to the temperature.

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

 

[Altis]

So if it ever runs below the max turbo speed (4.6 ghz) you would consider it thermal throttling? I guess i'm curious from what measurement are we using to say the clock speed has been reduced. Not trying to argue... just curious

Pretty much. Any time that it would run faster if the temperature was lower if a thermal throttle to me, in the literal sense anyways.

I know the turbo boost frequency generally isn't full-speed if it's doing it to a certain number of the cores. For example, IIRC my 4790k has base 4.0 GHz, turbo to 4.4 GHz but only if it's to 1-2 cores. If it's turboing the whole thing, it only goes to 4.2 GHz I think. It's a desktop CPU though so the turbo isn't going to be much since the base clock is already quite high.

It's definitely not a clear-cut thing. Chips are rated to a thermal design power (TDP), which is the maximum sustained power that it will use (and need to dissipate) at base clock frequency. Turbo boosting allows the CPU to go above this temporarily, so long as it stays within the thermal temperature limits. Laptop/ultrabook CPUs are basically a given that they won't be able to sustain full turbo boost because they aren't designed to run at that thermal level. However, with enough cooling, it can stay above base clock right up to its maximum frequency indefinitely.

The question is, would/should we say that it's thermally throttled if it's still running at/above TDP? I would say yes so long as temperature is holding the CPU back from higher speeds, as this is useful information -- you know the speed of the system can be increased if cooling can be improved. Laptop/ultrabook processors tend to have fairly low base clocks with quite a significant turbo boost, meaning low base power consumption and good sprints of high-frequency boosts when needed, but they cannot be sustained. An important factor in thermal throttling is not just if, but when -- if you're doing casual stuff, then you likely will get the benefit of the full speed in the short bursts that require it, without ever experiencing thermal throttling. But that same device might throttle for someone who's doing things like rendering videos over the course of hours.

If you had two laptops with identical CPUs, but one was able to cool it very well and sustain maximum frequency indefinitely, while the other could only do so for a few seconds before dropping to nearly base... would you say the one that throttles based on thermals is thermally throttled?

Anyways, that's how I see it. Sorry it's a bit wordy! Some interesting read over here at AnandTech.

 

[Spectrum]

From my opinion, if all cores are being pushed, the max capable for the i7 mini CPU is 4.3Ghz (intel specification). Thus if it is dropping to 3.5Hz to maintain temp below 100C then that is significant throttling. It is, however, still above the base clock.

On my 2011 mini, the all core maximum should be 2.6Ghz, but the maximum I ever get is 2.1Ghz (again slightly above the base speed of 2Ghz). In both cases, the thermal (or power?) constraints limit the all-core max turbo possible.

I'm interested to know if the 4 core i3 mini maintains its base clock of 3.6Ghz when all cores are pushed...it has no turbo, but 3.6Ghz is higher than the 6-core max that the i7 can maintain. But of course, the i7 is also doing 50% more work per minute.

I'd also be interested to hear what the max all-core turbo of the i5 model is (compared to the i7).

 

[LeeW]

So far with my new MM with i7 I have yet to see it thermal throttle, the fan curve could be better, the new cooling unit does work, when the fans ramp up it brings the temp down very quickly but lets it rise back to around the 90c mark when it could clearly keep it much cooler.

Sits quite happily at the top end of the clock speed for long periods (4 hours is all I have tested).

 

[Spectrum]

So far with my new MM with i7 I have yet to see it thermal throttle, the fan curve could be better, the new cooling unit does work, when the fans ramp up it brings the temp down very quickly but lets it rise back to around the 90c mark when it could clearly keep it much cooler.

Sits quite happily at the top end of the clock speed for long periods (4 hours is all I have tested).

If it is ever below 4.3Ghz...technically...it is throttling (for some reason - too much power draw, or too much heat). This doesn't mean the performance isn't excellent (it is), but it does mean it is below the maximum that the CPU is specified to achieve (by Intel).

 

[brentsg]

I've been tinkering with an i9-9900K in my desktop PC. If I run a CPU benchmark, or Handbrake, anything that loads the CPU to 100%, it immediately hits 4.7GHz on all 8 cores at 180W power consumption. It will hold this for ~30 seconds, then drop down to 4.0 on 8 cores @ 95W.

As an enthusiast machine, I can tinker with the BIOS to allow it to remain at 4.7 indefinitely if I wish, but it's putting a lot of strain on the VRMs. It never exceeds 85C, so I have thermal headroom.. so technically it's not thermally throttling.

I mention all of this because some people are expecting the Mini to hold all 6 cores at max turbo forever, which I think is unrealistic for such a small PC. Yeah my PC technically can do this if I want to, but the motherboard is way over designed, and the cooling system is nearly as large as a Mini. People are just expecting too much.

 

[Spectrum]

I've been tinkering with an i9-9900K in my desktop PC. If I run a CPU benchmark, or Handbrake, anything that loads the CPU to 100%, it immediately hits 4.7GHz on all 8 cores at 180W power consumption. It will hold this for ~30 seconds, then drop down to 4.0 on 8 cores @ 95W.

As an enthusiast machine, I can tinker with the BIOS to allow it to remain at 4.7 indefinitely if I wish, but it's putting a lot of strain on the VRMs. It never exceeds 85C, so I have thermal headroom.. so technically it's not thermally throttling.

I mention all of this because some people are expecting the Mini to hold all 6 cores at max turbo forever, which I think is unrealistic for such a small PC. Yeah my PC technically can do this if I want to, but the motherboard is way over designed, and the cooling system is nearly as large as a Mini. People are just expecting too much.

I completely agree with you - and thanks for sharing very useful information - particularly the insane power draw of this '95Watt' chip.

My original question was whether the max turbos are ever achieved in the mini though. From using with Intel Power Gadget (which may not be accurate), unless I am mistaken my 2011 mini doesn't ever reach the specced max turbo speed - even when temps are low. I wonder if this is because power draw is being limited.

If the 2018 mini can hit max speeds for short lengths of time (small jobs), then I think it is performing as advertised for the computer it is. If one wants the *fastest* 24/7 rendering/encoding computer, then, to me, this is not what the 2018 mini is designed for. A Mac Pro or an iMac Pro is a lot more suitable. However, the 2018 mini does make a very good (and cheap) mid-top end workstation (in my opinion).

 

[LeeW]

If it is ever below 4.3Ghz...technically...it is throttling (for some reason - too much power draw, or too much heat). This doesn't mean the performance isn't excellent (it is), but it does mean it is below the maximum that the CPU is specified to achieve (by Intel).

Agreed, that said this is not a machine you can reasonably expect to never suffer from thermal throttle, for the same reason I see it on the MBP, Dell, Surface Pro and so on. The form factor combined with ever more powerful hardware whilst providing benefit also creates drawback. My experience in any event is that it’s a minor issue at worst.

The thermal throttling issue is nothing new. What is new is the heightened level of fear mongering.

 

[Stephen.R]

TLDR: If you're buying a CPU expecting to have it run at 'turbo-boost' speed all the time, you're delusional. It's designed to provide short bursts of increased speed when the system is operating below its TDP, typically following periods of reduced activity.

Long version:

I don't agree with the idea "less than max turbo boost is the same as thermal throttling".

They're related, sure - but to claim that "my i7 didn't turbo boost completely, it's being throttled" is misleading.

Turbo boost is inherently about overclocking the processor above it's base frequency, when it's operating below TDP. If it's at TDP, turbo boost will not happen.

Throttling is about reducing the processor below base frequency when it's above TDP, to prevent damage. If you're above base frequency and go above TDP, and the frequency drops - that isn't throttling. It's the design behaviour of turbo boost - to use the accumulated 'energy budget' (literally their term, see the pdf at the end) from being below TDP.

If your computer is throttling and going below base frequency, it's a problem - this is why there was a 'bug fix' for the i9 processors that were being throttled a few months ago. The system should be able to operate at base frequency unimpeded.

Anything above that, via Turbo boost, is a cherry dependent on a number of conditions.

This is from Intel's 'why does my cpu throttle' page: (https://www.intel.com/content/www/us/en/support/articles/000029868.html?wapkw=throttle)

Processors have two modes of thermal protection, throttling and automatic shutdown. When a core exceeds the set throttle temperature, it will start to reduce power to bring the temperature back below that point. The throttle temperature can vary by processor and BIOS settings. If the conditions are such that throttling is unable to keep the temperature down, such as a thermal solution failure or incorrect assembly, the processor will automatically shut down to prevent permanent damage.

This is from Intel's Turbo Boost page (https://www.intel.com/content/www/u...ology/turbo-boost/turbo-boost-technology.html):

Intel® Turbo Boost Technology 2.01 accelerates processor and graphics performance for peak loads, automatically allowing processor cores to run faster than the rated operating frequency if they’re operating below power, current, and temperature specification limits.

Availability and frequency upside of Intel® Turbo Boost Technology 2.0 state depends upon a number of factors including, but not limited to, the following:

• Type of workload
• Number of active cores
• Estimated current consumption
• Estimated power consumption
• Processor temperature

When the processor is operating below these limits and the user's workload demands additional performance, the processor frequency will dynamically increase until the upper limit of frequency is reached.

Note: Intel® Turbo Boost Technology 2.0 allows the processor to operate at a power level that is higher than its TDP configuration and data sheet specified power for short durations to maximize performance.

This PDF describes the theory in more detail (page 14): http://www.hotchips.org/wp-content/...3.19.921.SandyBridge_Power_10-Rotem-Intel.pdf
[doublepost=1542373503][/doublepost]

What is new is the heightened level of fear mongering.

This much at least, I agree with.

 

[PJivan]

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

No as you can see in the link you provided you have the same issue with any manufacturer since the stock cooler is not enough, buy a dell tower desktop and you will see your i7 hit thermal limits, same can be said of almost any laptop in the market. Simply put, instead of limiting the frequency to a fixed number that can be sustained, intel decided to let frequincy run as high as TJ temperatures allow.
Why limit my frequency for short burst activities just because they could not be sustained for long? I mean this is what smartphone cpu do, this is what you do when you hit Moore's law.

 

[pl1984]

So my understanding is, the term thermal throttling means that the cpu will adjust its speed to below base frequency in order to lower the cpu temperature to protect itself from damage caused due to overheating.

This is thermal throttling.

However, all the reviews I’ve read of the i7 Mac mini are saying that it also suffers from thermal throttling. But their definition is different. I haven’t seen any reviews state that the cpu will go below base frequency, yes it reduces from the stated turbo freq but doesn’t go below base (3.2ghz). The reviews I’ve watched or read say the freq generally stabilises around 3.5ghz at around 90 - 100deg c. So this actually isn’t thermal throttling but performing above what it’s rated to.

This is not thermal throttling.

 

[Spectrum]

This is thermal throttling.


This is not thermal throttling.

Yet...if the cooling was better, the maximum stable turbo would be higher. Therefore, the processor is restrained ('throttled') by the temperature.

I think most people would be pretty upset if chips with large theoretical boosts could never achieve any kind of turbo boost after 30 seconds of work - yet maintained their base clock speed. Would you not argue they were throttled? (Yet according to the statements linked to above, they would be in specification and therefore not throttled.)

It would mean the i3 (3.6Ghz no turbo) would compute faster (except for the additional cores) than an i5 (3Ghz) or i7 (3.2Ghz).

I don't really mind. I just think the actual speed capability of the processors in the computer deployment of interest should be reported. Only stating "Turbo up to XXX Ghz" is only a small part of the picture. (As indeed is the CPU speed itself, I know.)

It would be trivial to perform a single/multi core "yes" stress test and report the stable final CPU speed(s). It would then give a prospective buyer the exact, maximum, CPU specs under single threaded and multi-threaded loads (for that deployment). Up to the buyer how they then choose to use (or not) that information.

Let's say a hypothetical MacPro launched with an identical processor to the i7 mac mini, but was 50% more costly because it had better cooling capable of letting a full multicore workload reach max boost speeds (4.3Ghz) indefinitely. Whereas the i7 mac mini stabilised at 3.5 Ghz. Wouldn't you be interested to know?

 

[PJivan]

It would mean the i3 (3.6Ghz no turbo) would compute faster (except for the additional cores) than an i5 (3Ghz) or i7 (3.2Ghz).

but this is impossible unless you use different coolers and they don't, if an i5 goes beyond JTmax at 3.4, then i3 would throttle to 3.4, there can be difference between same processor but you got my point.

Turbo boost is basically an automated overclocking capability that is absent on i3, you are not paying for the frequency (it's "up to XGhz" not "from xGhz") you are paying for the feature.

 

[pl1984]

Yet...if the cooling was better, the maximum stable turbo would be higher. Therefore, the processor is restrained ('throttled') by the temperature.

I think most people would be pretty upset if chips with large theoretical boosts could never achieve any kind of turbo boost after 30 seconds of work - yet maintained their base clock speed. Would you not argue they were throttled? (Yet according to the statements linked to above, they would be in specification and therefore not throttled.)

The idea behind turbo boost was to provide a means to balance single core speed versus multi core processing. If turbo boost could be maintained for all cores all the time then it would be the base frequency. Any processor operating 100% of the time at base frequency or higher is not thermally throttled.

 

[Spectrum]

The idea behind turbo boost was to provide a means to balance single core speed versus multi core processing. If turbo boost could be maintained for all cores all the time then it would be the base frequency. Any processor operating 100% of the time at base frequency or higher is not thermally throttled.

I disagree. The turbo is extra headroom that the CPU is capable of, at the expense of higher TDP than rated.
Put a crappy cooler in (below the TDP-spec) and even the base freq can't be maintained.
Put a mid cooler in (at the TDP-spec) and the base will be maintained (with bursts of boost possible).
Put a stupidly expensive over-engineered cooler in (above TDP-spec), and the max CPU boost speeds can be maintained.
It seems pretty simple.

Intel could (if they wanted to) simply rebadge the i7-8700B an "i9-120Watt" chip, and manufacturers would effectively have to improve cooling to manage the extra expected wattage and heat to prevent throttling. But it would still be the identical chip, but now have better performance...

Anyway...I don't need to argue this further. I don't really disagree with what people are saying. I'm just interested in the actual speeds capable.

 

[pl1984]

I disagree. The turbo is extra headroom that the CPU is capable of, at the expense of higher TDP than rated.

The idea behind turbo boost was to enable higher single core frequency versus lower multi core frequency within the same thermal envelope. You can disagree with this all you want but that will not make it any less true.

 

[PJivan]

Put a stupidly expensive over-engineered cooler in (above TDP-spec), and the max CPU boost speeds can be maintained.
It seems pretty simple.

It's called diminishing return. You can buy a 199 $ i5 8400 cpu and place a 65$ Noctua Cooler on top (1/3 of cpu cost) and you will get those 0.2-0.3Ghz more out of it at 60c or you can buy a much faster i5-8600K at the same budget, sure it will run hotter, sure you may not run it to it's own limit but still preferable. On laptop there is no way around it, bigger and heavier laptop is required.

 

[Stephen.R]

But it would still be the identical chip

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.

Turbo boost doesn't just enable whenever it can. It enables when the system needs higher performance.

 

[now i see it]

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.