Understanding Intel power limits?

[hat]

So Intel turbo, to my understanding, works by increasing the clock frequency as long as temps are fine and the processor isn't drawing too much power. But what exactly defines and enforces that control over power draw? My 2600k, a 95w chip, was constantly drawing 110w+ according to coretemp during my testing, and yet it never dropped below the speed I set (4GHz) nor did it throttle down due to power. It chugged along happily at 4GHz while constantly drawing 110w+ for hours, not seconds. I ran a quick test at 4.6GHz with a little overvolt to see what would happen, and the system never dropped below 4.6GHz or power throttled, despite even higher power draw (around 135w I think). I never changed any power settings either... so why is it these power limits evidently don't work?

 

[INSTG8R]

I believe the Tmax would dictate throttling over power draw. I could of course be wrong. I don’t pretend to understand Intel’s power setting with my previous Sandy and my current Haswell I’ve just used offsets to reign. In my boards overly generous voltage increases.

 

[kastriot]

Max core temp is 98C (Tjunction). Intel recomend a max of 72C for Tcase (the whole CPU)

 

[hat]

It's supposed to go by temps and power, though... and I haven't been able to make my chip throttle because of power. And it won't thermal throttle because I'm not into "suicide runs" and my cooler, while not fantastic by today's standards, can handle anything a reasonable person would do with this chip. And then we're seeing reports of ridiculous power draw on the 9900k. Where's the 95w limit at? Just to be clear: I'm not talking shit about the 9900k, just wondering why it's allowed to draw nearly 170w, seemingly ignoring those power limits?

 

[newtekie1]

I believe the Tmax would dictate throttling over power draw. I could of course be wrong. I don’t pretend to understand Intel’s power setting with my previous Sandy and my current Haswell I’ve just used offsets to reign. In my boards overly generous voltage increases.

I believe that was the case up until the latest generation.

Now the processor respects the power limit unless you adjust the power settings in the BIOS.

 

[hat]

I believe that was the case up until the latest generation.

Now the processor respects the power limit unless you adjust the power settings in the BIOS.

So those power draw settings were basically worthless? And now, with chips like the 9900k, they actually work? What if you don't want any of those options to work, so your chip can just draw whatever it wants? Usually there's fields like short term power duration, long term power duration, with finite values, so you can't really turn it off?

The thirst for knowledge on this one is great. I haven't felt like this since I first learned how to overclock, back when I had an AM2 setup...

 

[biffzinker]

yet it never dropped below the speed I set (4GHz) nor did it throttle down due to power. It chugged along happily at 4GHz while constantly drawing 110w+ for hours, not seconds. I ran a quick test at 4.6GHz with a little overvolt to see what would happen, and the system never dropped below 4.6GHz or power throttled, despite even higher power draw (around 135w I think).

Motherboard could or is automatically adjusting power limits when your manually setting the multiplier.

 

[hat]

Sure, but when the time duration is only a finite range, and usually for a small time (measured in seconds) how is it holding these power levels forever?

 

[unclewebb]

Many desktop boards decided to ignore the CPU TDP rating and instead set the turbo power limits sky high so they would not interfere with maximum performance. When overclocking, my Asus board sets both power limits to 4095 Watts.

If you want to see what your board is setting these limits to, check in the bios or better yet, run ThrottleStop. Click on the TPL button (Turbo Power Limits). If the bios has left your turbo power limits unlocked, you should be able to adjust these in real time and you can watch what happens to your CPU multiplier when it hits this limit.

 

[hat]

Many desktop boards decided to ignore the CPU TDP rating and instead set the turbo power limits sky high so they would not interfere with maximum performance. When overclocking, my Asus board sets both power limits to 4095 Watts.

If you want to see what your board is setting these limits to, check in the bios or better yet, run ThrottleStop. Click on the TPL button (Turbo Power Limits). If the bios has left your turbo power limits unlocked, you should be able to adjust these in real time and you can watch what happens to your CPU multiplier when it hits this limit.

I'm mostly curious what happens to newer systems, since it has been claimed that the new systems respect these limits, while older ones ignored them, or set them to impossible levels (you're not the first one to claim ridiculous settings like that 4095 watt limit).

This makes me wish I had a 9900k system, just to satisfy my curiosity here. I don't even care about upgrading, I just want to know how it works...

 

[cadaveca]

So Intel turbo, to my understanding, works by increasing the clock frequency as long as temps are fine and the processor isn't drawing too much power. But what exactly defines and enforces that control over power draw? My 2600k, a 95w chip, was constantly drawing 110w+ according to coretemp during my testing, and yet it never dropped below the speed I set (4GHz) nor did it throttle down due to power. It chugged along happily at 4GHz while constantly drawing 110w+ for hours, not seconds. I ran a quick test at 4.6GHz with a little overvolt to see what would happen, and the system never dropped below 4.6GHz or power throttled, despite even higher power draw (around 135w I think). I never changed any power settings either... so why is it these power limits evidently don't work?

The truth is that for years now, everyone has been overclocking with Turbo exclusively, but in order to do so, the locks have to be removed so that you can get the "most you can" out of your CPU. They've ALWAYS worked... but everyone turns them off without understanding or knowing or caring.

I'm mostly curious what happens to newer systems, since it has been claimed that the new systems respect these limits, while older ones ignored them, or set them to impossible levels (you're not the first one to claim ridiculous settings like that 4095 watt limit).

This makes me wish I had a 9900k system, just to satisfy my curiosity here. I don't even care about upgrading, I just want to know how it works...

They still do this. This is the BIOS engineers either playing the benchmarking game or simply tossing the locks off so we can tune things how we want.

Newer CPUs do offer more adjustments, for sure, but they could still use a bit more improvement for truly fine-tuned overclocking, especially considering the number of cores that we have at out disposal these days. The idea of overclocking in and of itself hasn't really evolved that much and most haven't been taking advantage of all these features that can tune things to really give you the most a CPU can offer, but it takes a lot of time to be able to get this just perfect. The perfect example of this is AMD's new Ryzen CPUs showing better benchmarks when things are left at "stock"...

...So I don't think a lot of people are going to have delved that deeply into that sort of thing. Intel's offered OC capabilities that you can tune to per-core loading, but most people just tune things to the same speed and then get stuck by the slowest core.

Sure, but when the time duration is only a finite range, and usually for a small time (measured in seconds) how is it holding these power levels forever?

Every time a task completes the CPU can "idle" for a short bit and then get going again. Clocks modulate all the time and load really isn't all that consistent, and that's why it is common to use simulated (fake) loads to test CPU stability. The time range is actually pretty long in that regard, and one short bit idle allows that timer to be reset.

Now we can set the maximum multi not just per load type, but we can also set it per core. So you can really have the potential to get a whole lot more out of a CPU that you could before. X299 CPUs have been able to do this for a while, and when you have 18-28 cores and a large die, you can even literally physically rotate the heat load around the chip to where you like.

With the 9900K... meh. Maybe someone else can show what it does. Anyone here got one?

 

[hat]

Tossing the locks off makes sense, and is how it should be once you enable overclocking features, I believe. Maybe that's why the notorious 9900k overheats often even just by enabling MCE... I just wish it was documented a little better. Those power limits make a lot of sense in average Joe's Dell tower, but not so much in an enthusiast's water cooled, custom built rig.

 

[cadaveca]

Tossing the locks off makes sense, and is how it should be once you enable overclocking features, I believe. Maybe that's why the notorious 9900k overheats often even just by enabling MCE... I just wish it was documented a little better. Those power limits make a lot of sense in average Joe's Dell tower, but not so much in an enthusiast's water cooled, custom built rig.

I agree, but it would be nice if there was a simple "switch" in most BIOSes to offer true "stock" settings of a CPU. This is specifically something I looked at a lot and kept an eye on in my motherboard reviews, but I did kind of feel like a small voice in a giant cave as to what might be really going on... that said, it's nice to see that some reviewers are paying more attention to these things, as the more than we are critical, the more things will improve.

 

[hat]

From what it sounds like to me so far, the moment you switch "A.I. Tuner" (or whatever option that allows overclocking) from Disabled to Enabled, locks come off. As some other articles I've read stated, I never had to worry about it with my old i5 2400, because no matter what I did it wouldn't draw 95w anyway. I guess it's coming into question more with chips like the 9900k that gobble power and even chips like the i5 8400 that have ridiculous boosts (2.8GHz normal frequency, 3.8GHz boost).

 

[MrGenius]

My question is...why are you confusing power consumption with TDP? They are not the same thing...at all. TDP is a measure of average heat dissipation in Watts(which IS NOT what you're observing). Power consumption in Watts is Volts x Amps(which IS what you're observing). Just because something is consuming X amount in Watts(V x A) of electrical energy, doesn't mean it's dissipating(or experiencing) that same X amount of Watts as heat.

TDP

Thermal Design Power (TDP) represents the average power, in watts, the processor dissipates when operating at Base Frequency with all cores active under an Intel-defined, high-complexity workload. Refer to Datasheet for thermal solution requirements.

https://en.wikipedia.org/wiki/Thermal_design_power
https://en.wikipedia.org/wiki/Power_rating

 

[unclewebb]

Intel CPUs constantly calculate an estimated power consumption value. This value is then compared to the long term and short term turbo power limits. This is how the CPU determines whether it should be using turbo boost or not. The turbo time variable is an approximation. After a fixed amount of time running at the higher power limit, the CPU will switch and be limited by the long term power limit.

Intel recommends that the long term turbo power limit be set equal to the rated TDP of the processor. The short term limit should be set approximately 25% higher than the rated TDP.

Perhaps that is where the confusion comes in. Intel themselves are using the TDP value in a different context to control the turbo boost feature of their CPUs.

The 2nd Gen all the way up to the latest 9900K all use the exact same register within the CPU to store the turbo power limit values and their speed is constantly adjusted based on these values. Lower these limits and all of these processors will throttle and slow down. Probably just a lot more noticeable now because the 9900K running AVX instructions is going to be a lot more power hungry. When the turbo limits are set to the rated TDP, you will see a lot more throttling compared to the 2nd Gen CPUs when their turbo limits are set to their rated TDP.

 

[John Naylor]

So Intel turbo, to my understanding, works by increasing the clock frequency as long as temps are fine and the processor isn't drawing too much power. But what exactly defines and enforces that control over power draw? My 2600k, a 95w chip, was constantly drawing 110w+ according to coretemp during my testing, and yet it never dropped below the speed I set (4GHz) nor did it throttle down due to power. It chugged along happily at 4GHz while constantly drawing 110w+ for hours, not seconds. I ran a quick test at 4.6GHz with a little overvolt to see what would happen, and the system never dropped below 4.6GHz or power throttled, despite even higher power draw (around 135w I think). I never changed any power settings either... so why is it these power limits evidently don't work?

My son's does, tho i can't remember how far .... back then there was a lot of talk about turning off Speed Step and C states to get an extra 0.1 Ghz or whatever... I never saw the sense of having the CPU run full tilt 24/7. The system Im typing from (4770k) peaks at 4.7 Ghz and settles down as low as 800 Mhz when not under load