CPU 45W TDP Limit

So I started an interesting discussion with @hmscott , he claimed that apparently there is a 45W TDP limit in his notebook, which I simply cannot experience in mine.

So I decided to make this thread in order to figure out if either hmscott does something wrong or if indeed some notebooks out there are indeed limited for some reason.

So if you would participate in this experiment do the following:
- Download XTU and setup your CPU
- Config ur CPU to good performance
- Run some kind of benchmark, XTU benchmark should be fine, monitor your TDP W.
- Also Monitor your Benchmark while having HWinfo64 on to see min and max wattage etc.
- Post a screenshot like this one below:
- Provide your notebook model


According to hmscott, it seems that his CPU gets locked to 45W TDP when his short turbo is running out, implying that his notebook is overwritting his limit which he set on turbo boost to 45Watts, which is interesting, since if that's the case that would mean that some notebooks restrict overclocking.

I hope to see some results!

 

Demystifying how Intel-based mobile Core i7 processors work: Clocks, throttling and thermal performance explained.
https://gist.github.com/Brainiarc7/fc7e24f8bbd0336b2cf796e89800e3e8
...
Second, you can run out of power (watts/TDP), referred to as power throttling. This is was usually as a result of an inefficient power adapter, being on battery, or simply the motherboard of the laptop not allowing enough power to be fed to the chip (either by a physical limitation, or by not allowing the turbo boost power limit to be increased), however to date all HQ chips (the only chips intel is using anymore) throttle to their default long power TDP (usually 45-47W) after a couple minutes (more on that below). If this is your issue and buying a bigger power adapter (assuming one is available) does not fix it, then buy a better notebook (or one of Clevo's desktop-CPU-using notebooks if you have a HQ chip).
...
Intel Core i7 49xxHQ/5950HQ variants: Increasing the TDP limits and amperage, etc via the BIOS or XTU DOES actually work for these chips. Confirmed using both Alienware and MSI notebooks. The problem however, is that the CPUs only keep the higher power limits for approximately 2 and a half minutes, just like with the 47xxHQ chips. This is more than enough for most benchmarks however, as anything that doesn't cross the base power consumption for the CPU will hold (such as GPU-heavy parts) and the parts that stress the CPU are usually ~1 minute long at most. When under serious load in gaming, however,these chips are not for you. They're not going to hold their clocks (likely not even stock, unless you seriously under-volt) under extended load times.
...
Intel Core i7 6700HQ chips: No clock adjustments are possible (even turning down turbo boost seems impossible in tested machines; you WILL have 3.5GHz 1-core, 3.3GHz 2-core, and 3.1GHz 3-core and 4-core turbo, or you disable turbo entirely). There is no manual adjustment possible and no free +200MHz overclock granted. I do not know whether it holds its power limits or not, as all laptops tested have either overheated before breaking the power limits, or simply couldn't hit the power limits whatsoever. I have no idea whether this CPU can hold over its in-borne TDP limit under load, and I suspect that I will both never know, and also that it will never matter; you're probably not going to draw enough from this slow chip.

Speaking of slow, remember that this chip is a 3.1 GHz 4-core turbo lightweight. Every single standard Haswell mobile SKUs, not including the ULV parts (I.E. ignore the 4702HQ and 4712HQ), if working properly, are at worst, equal to this chip. If you have one of those machines already, this is not an upgrade, and is most likely a downgrade, unless your current chip cannot hold its turbo clocks for some reason."

The Haswell test that you posted showing 2 minutes of high power draw is not long enough, you need to run for at least 5 minutes, and I'd like to see a 15 minute run.

And, see my last post to you from the closed thread for additional XTU graphing to show package TDP, thermal and power throttling.

 

just for reference on how a cpu turbo boost apply, formula at pag 70 from intel datasheet






etc. I will attach for convenience the pdf of 3rd and 4th Intel CPUs because I dont remember where to see online by intel www site.

3rd gen family Ivy bridge CPUs
https://drive.google.com/file/d/1QxBUqOvksXfYMQhQcSnt_PrOd5tbPgkO/view?usp=sharing
4th Gen cpus
https://drive.google.com/file/d/1WwZ1QnXbUzusC2ib8dGwB-1p7xkOE3jw/view?usp=sharing

 

Thanks for the read and link!
My theory is hereby confirmed then.

Guess u gotta find yourself a modded Bios to get it from stopping to nerf your CPU over time

However this is so expected from Alienware:

So from what I can gather from that nice post was the following:
- MQ models apparently have no problem going beyond the limit.
- HQ models do have this set limitation, however can be fixed with a modded BIOS
- XM/MX models do not have this limitation and can go beyond as well.
- 6700HQ is inferior to older ivy and sandy bridge CPU's due to hard nerf.

In a nutshell, manufacturers nerfing their CPU so that people have a reason to buy next gen. Also Alienware is the pinnacle of disapointment.

 

What notebook? P157SM iirc?

Protip: ignore what XTU says. It's unreliable. Go by Hwinfo or Throttlestop, especially for throttle and power limit reasons.

I notice you have cTDP there... that popped up when I installed the 3940XM in my P370EM. It wasn't there with the QM CPUs. It would keep setting a 81.25W short power and 65W long power limit, I have to override this every boot and resume with a XTUCli script setting them to 95W/90W which allows 4.5ghz.

Your package power (TDP) seems to be suspiciously close to 65W, too... and what are the other limit/throttle readouts on HWInfo? It looks like your CPU IS throttling for various reasons...

Stress tests that show you the work being done can show up throttling. 7zip benchmark is a weak one but consistently shows MIPS. Intel Burn Test is a cycling stress test, it'll show GFlops and will show up any power limit you got.

 

My CPU doesn't throttle (as seen on the image), it makes sense that it's only around 65W considering it's "only" 4ghz and the voltage is 1.065. Also the "throttle" you're refering to is a 2-3mhz dip caused by AMP limit. I only have it on 90, should have had 100. (as far as I know throttle means if it dips below clock speed or it thermal throttles because of thermal, so I think the term "unstable" is much better suited rather than throttle since it does seem to mean something different.)

HWinfo reports no throttle at all (hilariously enough XTU does report AMP throttle limit EDIT: Looking at power limit exceeded on my previous, It seems that AMP limit is meant with that, so HWinfo does report a limit problem on my 90AMP setting.)

I can make another test if you want to see a straight line with 100AMP limit

EDIT:
Here it is:


not a single mhz dropped, 100AMP limit did the trick

Also yes, it's my P157SM.

@hmscott
Ur article says 2.5 mins.... Well still here u got a 6min stresstest :x
I can tell you 100%, that this notebook does not limit my CPU in any way shape or form after any time, I play games like Mafia 3, Witcher 3, Dota 2 which all are more CPU heavy and not a single time did I see anything else than 3990 on my MSI afterburner On Screen Display, ever.

Also I really should fix the temps, looking at 90s makes me sad.

So if the article is correct, then this means that custom BIOS are a must for anyone who can overclock their CPU.

 

I don´t understand what the issue is, since the 45W TDP limit dos not apply to the i7-4940MX.

 

Thanks for point that out, it's been a long circle of trying to turn him back to the locked CPU's, that's the subject of the previous thread, he doesn't seem to realize the power boost and TDP limit details are different for locked and unlocked CPU's.

It's the 7700HQ that @Unhappy User had with his Turbo Boost Timeout Window set to 0.25 seconds and I suggested he use XTU to increase that to the maximum. I was hoping that would improve his CPU performance.

Then it just kinda spiraled down from there...

 

That's what I was wondering about, since hmscott claimed that mobilechips are limited to 45W TDP, thanks to his article it seems that this is actually a microcode related problem and is mostly affected by HQ models, meaning even MQ models do not seem to face this issue.

This is massively interesting since some people might have the situation where they consider either a HQ or an MQ chip. For instance @hmscott can Overclock his CPU rather nicely, but due to the microcode in his BIOS he is not able to use the full potential, which means that if he can get hold of a custom BIOS that fixes the microcode, he will be able to get rid of his limit set by the BIOS.

 

It's the HQ CPU TDP power limit, not the BIOS, although there are power limits there too, as it doesn't allow more than about 62w CPU package power for that laptop, even when setting the upper limits to 200w.

If you have an unlocked CPU, you don't have that HQ locked CPU power limit that's built into the HQ CPU. No matter how high you set the Power Boost or Short Power Boost, once that Turbo Boost Time Window expires, it drops down to the 45w or whatever TDP limit that HQ CPU has.

That's why @Unhappy User running a 7700HQ with 0.25 seconds on the Power Boost Time window makes no sense, you want to have the longest duration in seconds possible, because once it expires your power is limited in the HQ CPU to the TDP set by Intel, 45w in that case.

You were applying the unlocked CPU tuning to a locked HQ CPU, and that doesn't map over correctly.

 

The thing is you can still fix this with a microcode BIOS update, this is a quote from the article you posted:

Now the question is which processors are "unfixable" and which are "fixable".
And yes, you might be lucky and find a BIOS that fixes this for you getting your 4GHZ or whatever you would like to OC it to, since you do indeed have a 5950HQ

EDIT:

Oh cmon, you didn't either, why else would you ask me to make benchmarks longer and longer because you thought my 4940MX would set a limit over time like yours.

Locked CPU's also don't require more power than 45TDP tho, as stated even on stresstest I can't get beyond mid 30s TDP on a 6700HQ. So the limit of 45 would not affect my performance in any way shape or form, also how the window affects nothing but a system that wants to go beyond the limit but is holding itself back due to BIOS/CPU.

The explanation I gave you with the window on that thread was 100% correct, however if the CPU rewrites or doesn't accept your changes which you set in XTU, then obviously the time window will become important because the set max TDP which the CPU/BIOS has set has stock a higher turbo short power TDP than a normal Turbo TDP. Now it also makes sense why your CPU acts the way it does.

I assume that the "fix" the guy who made the post was, is actually just an unlimited time window, which would allow higher TDP, mabe even unlimited depending on what the CPU/BIOS has set.

 

correction: My MSI GT62VR has an i7-6700HQ.

 

This x1000

I'm assuming Prema BIOS is also installed - where the point is removing all the limits imposed by current limits, turbo short/long power & duration, etc etc etc by making them editable.

FYI here's some shots of various limits. Note that NONE of these showed up any limits on HWinfo or Throttlestop because you're on your own with figuring out what's happening with an Ivy Bridge heh

A QM with unlocked BIOS and turbo limits was nice, but an unlocked XM CPU with BIOS freedom is just a tinkerer's dream. I don't think I can go back to locked down stuff ever again.

The limit is how happy you are with how many volts you're shoving into the silicon and how hot you're willing to have it get, just like real overclocking.

x42 (4.2GHz) on the initial phase of Intel Burn Test:

then after some thermal throttling (lower multiplier) was starting:


x42 no limits: straight 42.00 average multi, through the whole test run.

85A current limit juuuust being tickled:

65A current limit:


x45 at 85A current limit: (note avg multi at x42 was 41.98; higher voltage at x45 wastes power)


here's x45 tickling the turbo short power limit because this happens immediately:

and again with raised turbo power limits, a flat 45.00:

 

They weren't talking about hacking the CPU limit, they were removing impediments to changing the power limits in the BIOS, so they could be raised, and used by the CPU - but the same "TIMEOUT" happens eventually, after a couple of minutes.

And, actually it's more "breakable" than "fixable", as the HQ CPU / BIOS is set to work "correctly" as shipped.

The cooling and power are designed to handle the HQ CPU as limited in the BIOS, and not as it could run recognizing the Turbo Boost Short Power Max increased power.

Freaking out over what something isn't is one of the weird things people fixate on in this hobby.

Once hacked or unlocked (not really), you don't gain much if anything anyway due to power supply and cooling design. You may have bought the wrong laptop if you are trying to break it to fix it.

Just enjoy HQ CPU's for what they are, tune them as possible: undervolt and increase the Turbo Boost Power Time Window to the maximum seconds + increase the Turbo Boost Short Power Max to some arbitrarily high value - like 200w, and enjoy

 

I will not argue about cooling, we all know without exception, notebook cooling is terrible. Modding is almost always needed. I'm also talking from a persepctive from a guy who would want to get "the most" out of his system. And as you pointed out, the EC firmware and PSU are also a factor, however, there are notebooks out there that do have an "oversized" PSU and have a rather big amount of extra wattage to play with, those mostly apply to older systems tho, newer system are restricted as can be.

@bennyg:
Oh my god 102c, never seen that before

From my experience higher AMP to lower Voltage ratio works most stable and is cooler than the other way around.

Impressive clock speeds man, how stable are they running over time?
Also arent those voltages a bit high for an ivy bridge? Seems rather massive to me.

3.6ghz into almost 1.2voltage is excessive, or are you not able to change to static voltage on ivy bridges?

 

Cool

It's got the same limitations as all HQ CPU's, so all the rest is the same.

 

It's just the way it is, there's no reason to jump through hoops and waste so many hours, weeks, months, years on getting the laptop to do 1% more performance, you'll soon forget it's a tool to accomplish so much more.

Tune for the low hanging fruit, whatever can be done quickly, and enjoy using the laptop for what you bought it for.

Perspective, and learning the limits, tuning for best performance without making a career out of it, it's a good thing...

And hard as you might try...

 

Set the amp limit way high, it's getting in the road. Since P = I * V, as you are clamping current (amps, I) and lowering Volts by undervolting, it's yet another form of Power limit (Watts) on the CPU.

Go run an AVX workload (handbrake, realbench, prime, linpack/Intel burn test etc) and see what is really limiting your CPU as it tries to draw +20W on top of what you see there.

 

In some cases I agree in others I don't. If I see potential in something, then I wanna get the most out of it, see it as my hobby

Some people enjoy getting the absolute best ouf of what they have, if u're an everyday normal guy using a notebook then yes, I agree fully, most people won't even fully use the "power" of a 6700HQ let alone some massively overclocked XXXX CPU, but for people who like to tinker with notebooks, I think it's important for them to know before buying an HQ processor, which can do what and what they can't.

Also that racoon is worth a like :'D

 

I was told that AVX workloads in prime break Haswell CPU's. I'll try intel burn test and see what happens.

 

4.2 is completely stable at stock voltage and safe for the clutzy auto fan profile. Even today as I was encoding a bunch of stuff (wrote a neat script to have multiple computers simultaneously work at encoding a large bunch of files in a network shared directory) and it was over 25C ambient inside.

No ability to alter volts other than add turbo voltage at the high end on ivy bridge. That is what the FIVR adds on Haswell onwards.

I don't leave fans on auto and run AVX benches at max OC on 30C+ days very often - and it was only for 20 seconds to generate those screenshots. You can only find some limits by exceeding them.

 

Just don't set to 4.0ghz first try.

The AVX workload doesn't break the CPU... the frequency you're trying to run it at can break a CPU you may think is "stable" at sane everyday-like workloads, because if you've tuned to the limit, you'll go way past that limit when you add all that extra power and heat.

 

I'll do some tests later today

thanks!

also before testing I should probably repaste, because at its current state I won't be doing it :x

 

No 45W/47W limit on this i7-4720HQ CPU with unlocked firmware.

Here's a screenshot of it averaging 55W over a 4-minute wPrime 1024M run:

Spoiler

Here is Prime95 Small FFTs (AVX) for 5 minutes at 65W:

Spoiler

^That was with TDP locked at 65W, so multiplier sat at 30x or so. We all know how AVX automatically overvolts and massively increases power consumption on Haswell. The CPU hits almost 90W if running that same test with unlocked TDP at 36x multi, reaching critical temperatures rather quickly on this notebook:

Spoiler

However, with a -50mV undervolt on core and cache, the CPU can run non-AVX Prime95 Small FFTs indefinitely at 36x, sitting in the mid-50W range at low 80C's:

Spoiler

These are the CPU power limits, set in BIOS, which I basically copied from a desktop 4770K:

Spoiler

 

As I said, it's thermal throttling and overheating - pushing way past the cooling design limits, and probably stressing the power delivery design.

It's not designed to do that.

How much more work do you get done by having done that mod? Do you run long batch jobs in shorter time, and have more turn over in them. Is there any real gain in your day to day use?

What are your daily running settings, what you set things to when not benchmarking - to extend the life by reducing the thermal and power stress to the system? I hope you don't run limitless extreme's all the time.

Or do you only gain posting benchmarks once in a while? How often can you even do that?, not many people are going to know enough to be impressed. It's very nice.

It's fun I'm sure to cheat the design, but really what does it accomplish?

And, again, that's the Haswell firmware microcode hack, I haven't heard of anyone cracking through the 6700HQ or 7700HQ microcode, have they?

We were originally talking about giving somone advice on how to tune their locked HQ CPU through XTU / Throttlestop to get best performance, but I assumed not by making their laptop into a little Chernobyl that they would have to tweak and tune appropriately for every load, or have 100% fans shrieking in their ears while using it. Which isn't even possible for a 6700HQ / 7700HQ.

I made one simple comment after looking at his XTU settings:

"I don't see anyone else mentioning it, so you might want to check how high you can set the Turbo Boost Power Time Window - currently at 0.25sec, it should be at least 8 seconds, but see how high you can set it in that drop down, I've seen up to 28 seconds, but not sure what the 7700HQ is limited to.

That's why your CPU keeps dropping Turbo Power, it's only keeping it at full Turbo Short Power for .25 seconds "

It's what someone with a non-Haswell HQ CPU - or even a Haswell CPU they don't want to Chernobyl - and a tunable Turbo Boost Power Time Window would want to hear - their Window setting is too short - check the drop down menu and see if you can change it to the maximum available setting.

@Unhappy User - have you tried that?

I now think the 6700HQ / 7700HQ won't actually benefit from that change - but it won't hurt it - because the power drawn under load seems to never approach that limit.

In my previous unhacked Haswell / Broadwell use the CPU's would at least be tuneable to draw more than 45w for up to 28 seconds, but the 6700HQ / 7700HQ don't draw that much power even under the most strenuous CPU load...

So all of these posts just to ask a guy to try a setting have been wasted, he could have simply tried it and told us it made no difference

The poor Mod's thought this was going to turn into some kind of BGA bashing, but actually we've all been pretty well behaved and not done that, it's just a fun walk through discovery of what the limits are, and why it's a waste of time to hack them for day to day operation. The design won't take the added power / thermal load enough to make the hack useful.

Sorry if I am a bit down on over-driving hardware past it's power and cooling design limits. I started my career as an electronics repair technician for large analog / digital design companies in High School and College, and I've seen what accidental over power draws do to the hardware I have to repair. Ever since those years of hands-on experience I know that the design limits are there for a good reason, so I always work within the limits of the design and I don't think it wise to exceed them as I've seen the damage it can do.

All that danger to your expensive hardware, just to get bragging rights that you broke your device such that it can briefly gain the performance of another design, never made sense to me

To me going past the limits useful for daily use makes no sense for the majority of users looking for performance and tuning advice, they don't want to go through a bunch of axles trying to use that modded performance:

Everyone loves watching guys blow their stuff up from abusing it, so that's another thing to keep in mind. If you don't want your stuff to blow up, don't take performance modding advice from guys with boxes and boxes of broken axles sitting in the garage.

 

It was just to show what's possible with an HQ CPU when it's freed from the shackles of stock cancer firmware. Like I said, under real-world max loads, i.e. not when running Prime95 AVX Small FFTs, TDP tops out in the mid-50W range and thermals are managed just fine by this laptop's cooling.

Yes. In any workload that stresses the CPU at 100% for more than a minute or two, I can actually run the CPU at its full rated speed, instead of throttling.

Those are my regular settings that I showed. 38/37/36/36 1/2/3/4/-core multis, 84W/105W long/short turbo power limts, 8s time limit, 95A current limit, -50mV core and cache. Like I said, basically copy-pasted from stock 4770K settings, so hardly extreme.

I benchmark after physical hardware changes, when testing overclocks, after driver updates, and if I suspect performance degradation.

Aside from that, I don't benchmark an already tuned system often.

"Cheat the design" is a funny way of looking at it. I call it getting the performance I paid for. But we probably won't see eye to eye on this.

I'm not running the old bugged microcode ( see here), otherwise I'd be able to overclock past 3.6GHz, which I can't.

 

I hacked my ‘un-overclockable’ i5 6500 to hit 4.3GHz and outperformed my old i7 2600K

 

But, this BCLK OC hack doesn't unlock power limits... so you are still limited by 45w power limit after the Turbo Boost Power Time Window expires, right?

A 2700K I built in 2011 is still running at 5.0ghz, probably outperforming this hack, running cool with Noctua air cooling, a bunch of Noctua fans, nearing 7 years on AS5

IDK, why not just buy a full on OC'able K CPU for that Z170 motherboard, it seems like a lot of untapped potential left for that build.

"With that done, I’ve simply boosted the CPU base clock and DRAM frequencies, and added that extra 0.35V in the CPU voltage settings. Since I have mentioned voltage twice now and it is making me nervous, please note that neither I nor PCGamesN are responsible for you frying your hardware by changing those settings. This, and whatever fire you start, is on you."

Comforting Words...

That reminds me of this cool hack, putting a Suzuki Hayabusa Hyperbike Engine in an Austin Mini Cooper

He put a HAYABUSA Motor in his MINI - It’s SICK!


I preferred my Hayabusa Engine in it's original skins, up to 180 MPH

Suzuki Hayabusa GSX1300R - Acceleration 0-300km/h & Startup & Exhaust Sound & Burnout & Dyno

 

Thanks for your post, very useful!

So even older haswell HQ CPU's are limitless when doing a microcode update. With a mod on the heatsink I'm sure u could somewhat get temps more "proper", but then again, stresstest makes the CPU way hotter than "normal" use anyways, so I'm sure your temps on gaming and stuff are rather low.

 

I didn't know Dell got own special designed unlocked Core i7 BGA versions from Intel for use in their notebooks. Damn, nice... 7820hk Extreme Edition and Intel® Turbo Boost Max Technology 3.0


And I didn't know Core i5 processors come with Intel® Turbo Boost Max Technology 3.0. I really need to step up and read me up more on newer tech

 

Technically it's a microcode downgrade. Stock Haswell silicon has a bug where it ignores fused multiplier limits and allows things like this to happen on microcodes <= 0x02.



However there are some major caveats:
BGA i7s rarely honor TDP limit settings, almost always reverting to fused TDP values. Powercut is a godsend to this...
As expected running the processor like this makes it really hot, and as a result my actual "daily" multis are 44/44/41/41.
However for longer load sessions (like some intensive games, video rendering,electronics simulation) I need to drop the multi to 38x or else I'll run into PROCHOT.
This is on a Lenovo Y50, and using Conductonaut.

I have recently acquired a Precision M4700 for a decent price ($250-ish). It originally came with an i7-3720QM, which while decent, it lacked the horsepower I need.
Therefore when I saw a 3940XM for a low price ($150), I grabbed it and replaced the original processor.
Both processors did allow for TDP limits to be exceeded, will post pics later.

 

I'm not using the bugged early microcode that lets you increase multiplier beyond stock limits:



Yeah CPU temps are fine when gaming (60-70C) or encoding long videos (80C).

 

Hi guys Im having Power limit throttle issues and maybe you guys can help. It happens when TDP package passes the limits that are locked. I want to unlock them but my BIOS is super simple and a can tweak almost nothing. Using HWinfo32 shows:

CPU Thermal Design Power (TDP): 47.0 W
CPU Thermal Design Current (TDC): 85.0 A
CPU Power Limits (Max): Power = Unlimited, Time = Unlimited
CPU Power Limit 1 - Long Duration: Power = 47.00 W, Time = 28.00 sec [Locked]
CPU Power Limit 2 - Short Duration: Power = 58.75 W, Time = 2.44 ms [Locked]

So i can set the power limits to an unlimited value, but it seems the manufacturer locked the limits.
How can i unlock them??!!!

Thanks,

I have a clevo n170sd:
i7-4720HQ
GTX960M
8GB DDR3 1600mhz
windows 10 pro
AC 120W