Haswell-ULT: throttleing by design?

It's been the 2nd time I came across a review on notebookcheck.com where a Haswell i5-U / i7-U throttles although the CPU stays within it's thermal design:
Test Toshiba Portege Z30-A-12U Notebook - Notebookcheck.com Tests
Test Samsung ATIV Book 9 900X3G-K01DE Notebook - Notebookcheck.com Tests

In both cases the CPU locks at 1200MHz. The reviewer refers to the low TDP (15W) to be the cause, and it sounds like he is speaking of a feature of those CPUs.
I'm wondering if this can be true!???



My Ivy bridge i7-U doesn't show this behavior and runs forever in turbo mode.

 

Yeah, it's true (it's not the cpu so much - it's the implementation that the notebook manufacturer does with it).

Check out the ThrottleStop thread (last couple of pages) for unclewebb's take on that issue.

 

Can't fight physics, if you lower the tdp you lower the max frequency.

 

Exactly what tiller said; my Surface Pro 2 does the same with it's i5-4300U.... temps max at 80C per Microsoft's design by beginning throttling at that point. I'm sure the CPU could go much further but for design reasons regarding other components in such a small chassis, it is what it is! Fortunately, at least in my SP2 scenario where I lose a bit of boost, it only drops a few hundred MHz, not all the way to 1200... that's a heck of a drop in those machines!

 

Anandtech noted throttling to stay within TDP in an article looking at Intel's recent graphics performance. They discovered that standard-power processors had no throttling problems, but the low-power ones (U- and Y-series) are pretty quick to throttle if you load the CPU and onboard graphics at the same time.

 

Interesting if the new Haswell ULV parts are throttling, maybe the 15W limit is just too low. I have the older Ivy Bridge ULV with 17W TDP and even with working IGP the two CPU cores can run full speed all the time. I think it is also possible the Haswell architecture has worse performance/watt than Ivy Bridge and that is gives an another reason to throttle. However do not forget guys how badly the AMD APUs are throttling, even the 35W versions do badly and the under 20W parts are disaster.

 

Yeah it kind of sucks. My Sony Vaio Pro 11 can actually perform decently but the iGPU throttles like crazy, so while it could game, there can be massive FPS spikes, and the CPU isn't even that hot, less than 70C. It seems to me that they way underrated the TDP. The Intel "U" CPU's should probably be more like 25W than 15W to perform as needed.

 

An earlier post said the limits can be set by the laptop manufacturer as well and so if the cooling solution is small like a tablet or ultra thin, they can opt to limit the turbo more then other laptops with the same chip. This really messes with benchmarks from device to device cause the same chips may have stricter performance settings on one device.

 

Not so much is a bit vague. So is the TDP too low for the CPU to maintain it's advertised frequency or not? The temperatures in the above two examples are obviously fine

If the manufacturers artificially cripple the CPU w/o reason, I'd feel pretty cheated.

Btw, one of the above throttled laptops is actually the Haswell successor of my Ivy-Bridge machine (same chassis). So due to 2W TDP difference I have to go all the way down from 3Ghz to 1.2Ghz? Is this the new way of upgrading?

If you can't fight physics you better don't advertise an illusion.

 

You're not reading what is written: the cpu is capable of the spec's it promises. The manufacturers though can manipulate some parameters to fit it inside a less than optimal chassis.

This is a case of buyer beware (as usual) and if you're feeling cheated; the issue is lack of researching - not lack of available information.

Even with no information available, that is what the return period is for: if it doesn't surpass your previous setups performance; return it. No matter what the advertising promised.

 

Companies advertise illusions all the time. Tell me the last time you saw a realistic beer commercial.

As usual, the devil is in the details. The processor can definitely achieve its rated turbo speed, but Intel makes no promise of how long it can stay there. And when both the processor and the video parts of the same chip are trying to use all the available thermal power at once, something has to give. That tends to be clock speed. So you can say that Intel or your notebook manufacturer advertised an illusion, but they didn't actually lie about the capabilities of the product. And although it may be dishonest, it's not illegal to leave out information as long as that information does not present a danger.

 

You don't seem to be aware of what you have written yourself. By saying "it's not the cpu so much" you are implying it does relate to the CPU itself to a certain extent, regardless of the statement that follows.

Yes, unfortunately notebook manufacturers cripple the CPU. The Windows drivers of my machine eg have another useless feature, ummm restriction, which throttles the CPU when running on battery or when in 'silent' mode.

Just because they do it, does not mean it was necessary to implement. And in the 2 given examples in the OP this is not a chassis (temperature) issue. There is simply no reason to throttle, but they do.

A lack of researching? Are you serious? People on nbr might research such a thing. 90% of customers don't. And the majority of them will even never find out about being fooled.

 

If I'd buy a beer with 5% vol advertised and it delivered only 2% I'd be pretty pissed... Well actually not so pissed

 

There's a reason there's a published rating for a CPU. It's the rated speed your CPU is guaranteed to run at. Everything above that is boost and it's gravy. This was the concern from day one with Sandy Bridge when "boost" was implemented and user overclocking was removed. Granted they've relaxed a bit and do offer some user flexibility in this regard, but you will still need a higher end CPU (like quad core i7) to take advantage of it.

 

oled, you're reading my statements wrong (or I'm writing them wrong) - either way, that is not what I meant: The following statements do modify and clarify anything made previously (i.e. read again).

 

Yes they do, read >here<, basically

Lots of other reasons for throttling though but those are the primary ones for the CPU on Intels part.

 

You misunderstand. Intel definitely has specifications that indicate the conditions under which turbo can exist, but for a typical end user, those conditions do not correlate with a known workload. To rephrase my previous comment, Intel makes no promises regarding whether the CPU will remain at turbo frequencies for the entire duration of whatever program(s) happen(s) to be running at the time. And that specifically seems to be the main complaint that started the thread. The higher the TDP of the processor, the less likely it is to have this problem where it drops to lower or minimum frequency in order to preserve rated TDP.

As the thread to which you linked was specifically talking about desktop processors, which are rated at much higher TDP values and therefore do not have these problems, it did not truly address the statements in this thread.

Self-serving edit: post #1000!!

 

Principle is the same for both desktop and laptop CPU's. Change the settings of the laptop CPU and it will run full turbo always providing the laptop thermals and power delivery can handle it.

 

Check out this thread on the HP Spectre 13t-3000 Ultrabook with i5-4200U/HD440 or i7-4500U/HD4400 an(huge thread, you'll need to read a lot; I believe a lot of the talk of throttling is about 1/3 of the way in. Or try the search function in the thread, it works far better now than in older versions of NBR.

To get to the point, on this particular model, people found two things to keep the cpu in the high 2's most of the time:

1) uninstall Intel Collaborative Processor Performance Control (CPPC) driver. Without going into depth about what CPPC does or how it does it (in part because I don't fully understand it!) it is an "energy management" add-on, and numerous owners (including yours truly) tested cpu throttling/turbo retained GHz with and without CPPC and found that uninstalling it (in some models you need to install it from the Intel site to have it, but I believe the majority of Haswell UBs come with it installed) restored full turbo-ing and sustained clock speeds above 2.5Ghz and often over 2.9Ghz for sustained time when load required it. Others may have comments as to how/why this works - some may even claim it doesn't, as Intel does in its propaganda, but trust me, on the HP Spectre 13t, it virtually eliminates the problem - except when running on battery, in which case the model caps cpu clock speed at 1.6Ghz.

2) As recommended earlier, check out Throttlestop.

It seems clear that there are three identifiable causes of cpu speed reduction (not including reducing maximum speed in power profiles): heat that can't be controlled by ultra thin cases and weak fans; the presence of Intel CPPC; possible throttling programmed by the mfr well before heat limits are reached. The Spectre 13t is about as thin as any ultrabook around and it is all metal, which means it is a conductor of heat, though I've read that this can also act as a heat sink, depending on the interior architecture of the computer.

Sorry for the somewhat disjointed discussion above. I have to say that, with all the theories and suggestions, nothing worked as well as uninstalling Intel CPPC (and disabling it in Power settings, though I assume the Power settings are inoperative if the software is uninstalled.

Good luck, and do report in. And consider specific models and what you can find written in threads of those models here.

 

This just doesn't seem to be the case with the ULV CPU's though. As much as I've fiddled with laptops and CPU's and performance, The ULV just doesn't hold up like its bigger higher voltage/higher TDP brothers. You can't change the settings of the ULV laptop CPU in any way, shape, or form, they're locked down. I can run at a cool 70C, and it will still throttle. AMD ULV parts exhibit the same behavior. It's like they forced the TDP rating even though it wasn't capable of sustaining performance at that TDP.

 

Which Haswell Ultra do you have?

For instance with Sony Pro 11 SVP11213CXS BIOS V710440B all the following can be changed.

Code:

Setting				Default	
========================================
Hyperthreading			Enabled
Active Processor Cores		All
Overclocking lock		Disabled
Turbo Mode			Enabled
Energy Performance		Performance
Package power limit lock	Enabled
Cpu Power Limits		0
Cpu Power Limit  Times		0
Cpu Power Limit3 Duty Cycle	0
DDR Power Limits		0
Platform power limit lock	Enabled
Core Ratio Limits		0
VR Current value lock		Enabled
VR Current value		0
ACPI CTDP BIOS			Enabled
Configurable TDP Level		TDP NOMINAL
Config TDP LOCK			Disabled
TCC Activation Offset		6
Passive Trip Point		95C

Even then it's not so important as long as there is enough current and thermals can be kept in check without over taxing the laptop power delivery spec.

 

The CPU product from Intel is capable of whatever it's advertised to do.

The cooling from the ultra book manufacturer is usually not.

 

I'm not sure where you see that? Please enlighten me. Here's BIOS and Intel XTU in my SVP11213CXB. It's locked down.

Spoiler

Not entirely. Why does it throttle at 70C even then?

 

It comes down to your definition of locked down.

Level 1. The option is not shown in the BIOS setup screen but is part of the setup and default is locked. This can usually be overcome through the UEFI shell or if running GPT boot through windows. An option of using GPT and Windows not realized by many.

Level 2. It's not part of the BIOS setup and is locked by BIOS. Requires BIOS modification

Level 3. Intel have locked the option. Hard to do much with this unless a bug is found, a loop hole or perhaps knowledge of microcode which is one of Intels tighter secrets.

For instance in the above post "package power lock" can be disabled by level 1, for an i5-4200U should not be a problem but perhaps with another processor TDP/TDC itself is locked by Intel which would require level 3.

 

If it's not adjustable with Intel XTU it's not available. And not part of UEFI shell either. It's simply locked down by Intel. I have the i5-4200U.

 

The UEFI shell has all the tools necessary to change the settings. The settings posted above come under level 1 and are therefore changeable as stated in original post, not level 3 (Intel locked).


Here's a post dug up where TANWare used Level 1 the UEFI Shell to enable extra bins on his i7-3820. Options for this were not adjustable from the user BIOS utility or adjustable in XTU.

I can understand why laptop manufacturers do not show a lot of the hidden options, seems akin to opening Pandora's Box. The less options provided the less chance of someone contacting them because they can no longer boot due to changing a BIOS setting without realizing what the consequences are.

 

at what speed should the IGP on an i5 4300u run at?

 

See:
ARK | Intel® Core™ i5-4300U Processor (3M Cache, up to 2.90 GHz)

 

I have a Samsung NP300U1a with an ULV 1.3 GHz i3 Sandybridge, and Samsung had an ingenious idea of ensuring long battery life.

They hard-coded the BIOS to lock the ULV i3 at 800 MHz upon entering battery mode, even when I'm running Matlab in a remote desktop (VMware) connection, much to my annoyance when running Matlab codes.

And the laptop is apparently incompatible with Samsung's own Easymanager software that might've allowed me to undo the battery mode throttling.

 

Have you tried Throttlestop?

 

This is not a BIOS lock but Samsung (Easy-) Settings at work. This is at least true for the Ivy Bridge Series 9. Running Linux I don't suffer those Samsung Windows driver restrictions.

Anyway, as djembe said, throttlestop can unlock the CPU.

 

You are absolutely right!

 

See how well the 4550U in my X1 Carbon performes with tdp above 20.
http://forum.notebookreview.com/thinkpad-edge-x1-x100e-x120e-sl-l/768535-x1-carbon-haswell-performance-boost-due-unlocked-tdp.html
Shame it's not permanent...

 

I wonder how you did this...