Retail Versions of Intel Core i ?Haswell? Are ?Hotter and Slower? Than Expected

[ajaidev]

"Although Intel Core i-series ?Haswell? microprocessors have a number of advantages over predecessors when it comes to overclocking, retail versions of the chips are not as good overclockers as pre-production versions of the chip. According to manufacturers of factory overclocked PCs, commercial versions of Intel Core i7-4770K cannot remain stable at speeds achievable by samples of the product. Intel admits: overclocking results are not guaranteed.""

http://www.xbitlabs.com/news/cpu/dis...ed_Report.html

haha told you ppl

http://www.xtremesystems.org/forums/...=1#post5191357

http://www.xtremesystems.org/forums/...=1#post5191376

If u dont get it more power consumption = more heat

[Rob94hawk]

Interesting. Just picked up the 4770k at microcenter yesterday and waiting for my mobo to get here. Either way we're going to hook it up to the Swiftech H220 and see how it goes. Will post with batch # soon.

[kl0012]

If u dont get it more power consumption = more heat

This "formula" is completely wrong. RTFM

[Eldonko]

Not sure I believe it. Curious to see some input from those with retail chips.

[informal]

We need a bigger sample of CPUs (from retail) to say for sure this is the case .

[The Stilt]

Intel most likely picked parts with lower leakage properties as ES parts to be sent to the partners and "special persons".
Even Intel's superior 22nm process has variations in leakage between the different parts.

This is not a big deal really as everyone does it.
Actually it is a very sensible thing to do, while making any conclusions based on ES parts is not

[xdan]

I think it might be true...i know 3-4 examples of retail bought in different countrys..... 5ghz 1.5v just to boot, 1 cpu couldn't boot at 5ghz on air.
Usually at lauch there were pretty good chips , more good than average.
Sandy Bridge launch, Ivy Bridge launch.
Haswell is seems that there few chips that can do 4.5-4.6ghz stable 24/7. Many 4.3-4.4ghz.

[WeaponX]

I'm currently testing a 4.6ghz overclock at 1.325v. At 1.3v it was almost completely stable with temps hitting 78c at it's hottest core. Ran 2 hours of Prime 95 and played 2 hours of COD Black Ops 2 just fine. Then BSOD over night. So I bumped up the vcore just a little and doing some more testing. So far it's much better than my 3770K I had. My specs are in my sig.

[Mechanical Man]

This "formula" is completely wrong. RTFM

Wtf u smoking. Ofc more power consumption = more output heat when talking about microprocessors. Or do you think that used power somehow vanishes in thin air?

[Particle]

This "formula" is completely wrong. RTFM

I'm tired of seeing this. A CPU does not emit meaningful amounts of EM energy (radio, visible light, etc) nor does it emit sound or move. Virtually all power consumed by the processor is converted into heat energy. More power = more heat. Heat isn't the same thing as temperature (heat is volume, temperature is density), however, so perhaps that is where the confusion lies.

[ajaidev]

This "formula" is completely wrong. RTFM

Its correct mostly

P=CV2f

Some notes on it from Intel

http://software.intel.com/en-us/blog...o-obvious-pt-1

http://software.intel.com/en-us/blog...obvious-pt-2-2

You can treat the notes as manual.

[LordEC911]

Intel most likely picked parts with lower leakage properties as ES parts to be sent to the partners and "special persons".
Even Intel's superior 22nm process has variations in leakage between the different parts.

This is not a big deal really as everyone does it.
Actually it is a very sensible thing to do, while making any conclusions based on ES parts is not

Higher leakage = better overclocks if you have ample cooling.
Most reviews seem to have the "highend" closed loop watercooling products to overclock with.

[The Stilt]

Higher leakage = better overclocks if you have ample cooling.

Yes and no.
In many cases this is true, however a close loop watercooling is not something I would call "ample".
The leakage decreases along with the temperature, however the real drop happens in temperatures well below zero.

If high leakage properties in semiconductors would be something desireable why are the manufacturers spending millions to fight against it?

[LordEC911]

Yes and no.
In many cases this is true, however a close loop watercooling is not something I would call "ample".
The leakage decreases along with the temperature, however the real drop happens in temperatures well below zero.

If high leakage properties in semiconductors would be something desireable why are the manufacturers spending millions to fight against it?

I understand all that.
A highend closed loop is going to be quite a bit better than the stock cooler that reviewers used to overclock with or even the old highend aircoolers.

Because high leakage causes high power consumption, as I'm sure you know, which doesn't fit with companies trying to hit specific TDPs.
So it isn't desirable when you are trying to bin tens of thousands of high leakage chips but on the flip side, it is highly desirable for overclockers.

[The Stilt]

Leakage does not raise the power consumption directly.

E.G.

TDP = 100W

Rated Core VDD (low leak): 1.30000V -> IDD = 76.92A
Rated Core VDD (high leak): 1.22500V -> IDD = 81.63A

Parts with high leakage properties run hotter and the heat will increase the leakage even further.
Until a certain point it is manageable but the "mad mans loop" is still present from the very beginning.
This is exactly the reason why high leakage parts generally do better under extreme cooling.
High leakage parts are considered as lower quality parts, not higher.

[paulbagz]

For non LN2 OC is it that big of a deal if it can only hit 4.5Ghz?...

-PB

[Levish]

Could be hotter due to on chip additional circuitry that was previously on board and better gpu.

The two retail ivy chips I tried did 4.5 - 4.6 at "24x7" clocks and volts a 5Ghz Haswell would be a nice step up, ~10% improved OC and another 5% from IPC improvements.

[DaKon]

Has anyone started an OC thread on the 4770k? I would like to add my two bits in, when I get my motherboard next week. If we can get a broad sampling of retail, it would shed light.

[Rob94hawk]

Has anyone started an OC thread on the 4770k? I would like to add my two bits in, when I get my motherboard next week. If we can get a broad sampling of retail, it would shed light.

http://www.xtremesystems.org/forums/...nd-o-c-results

[LordEC911]

Leakage does not raise the power consumption directly.

E.G.

TDP = 100W

Rated Core VDD (low leak): 1.30000V -> IDD = 76.92A
Rated Core VDD (high leak): 1.22500V -> IDD = 81.63A

Parts with high leakage properties run hotter and the heat will increase the leakage even further.
Until a certain point it is manageable but the "mad mans loop" is still present from the very beginning.
This is exactly the reason why high leakage parts generally do better under extreme cooling.
High leakage parts are considered as lower quality parts, not higher.

We will just have to agree to disagree then.

[onewingedangel]

Haswell seems to consume less power but run hotter, so it is less likely to be purely leakage, and more of a power density issue. Lower power consumption/leakage parts will run cooler but still be comparatively warm. Parts running hot is not a problem as long as they run within specs and Haswell absolutely does.

You also need to factor in that Intel is currently binning the lowest leakage parts for the performance notebook market, so it is inevitable that the higher leakage product will be sold as desktop parts, and most likely to be sold in trays to OEMs that are less likely to overclock beyond rated specs. Those expecting more than they paid for really have no-one to blame but themselves for their disappointment.

It would be interesting to see how retail packaged chips that you and I can buy compare to the tray CPU's the boutique builders are complaining about - this comparison may show the thread title to be misleading.

[Hornet331]

Those expecting more than they paid for really have no-one to blame but themselves for their disappointment.

Thats pretty much the key phrase here. Intel has no obligation to provide anything on there products that are out of there set specification.

[st0ned]

I understand all that.
A highend closed loop is going to be quite a bit better than the stock cooler that reviewers used to overclock with or even the old highend aircoolers.

Because high leakage causes high power consumption, as I'm sure you know, which doesn't fit with companies trying to hit specific TDPs.
So it isn't desirable when you are trying to bin tens of thousands of high leakage chips but on the flip side, it is highly desirable for overclockers.

That's not right. As long as you are not going too cold high leakage will always be worst. As a matter of fact leakage = bad. That's why hardware manufacturers have developed a handful of ways to reduce leakage.

Leakage changes with both voltage and temperature, it increases/decreases with both. You can see this paper about it.

See more explanations below.

Leakage does not raise the power consumption directly.

E.G.

TDP = 100W

Rated Core VDD (low leak): 1.30000V -> IDD = 76.92A
Rated Core VDD (high leak): 1.22500V -> IDD = 81.63A

Parts with high leakage properties run hotter and the heat will increase the leakage even further.
Until a certain point it is manageable but the "mad mans loop" is still present from the very beginning.
This is exactly the reason why high leakage parts generally do better under extreme cooling.
High leakage parts are considered as lower quality parts, not higher.

More leakage means more power consumption, if you want to have X current from the Source to the Drain if say 10% of that current leaks, undesirably, thru the gate you have to feed the FET with ~X+10% . This is an extreme case only to be used as an example, leakage currents are much smaller in reality.

Why is Leakage bad ? Well for starters it will make your chip run hotter, on the other side with the rise of leakage you will, ultimately, destroy your semi-conductor. It can also be a cause for error due to insufficient current going thru the normal path, given a high amount is leaking thru the gate.

So why do extreme overclockers prefer high leakage chips ?

Well this isn't always true, IB prove this, when the best chips in cold were the ones doing 5GHz with ~1.2v and lower on air, so chips with low leakage. Well when you use LN2 two things happen, first you boost electric conductivity thus making signals move faster in the conductors, this is one cause for X frequency not being stable, signals take too long to propagate along the chip and some inconsistencies happen causing chip failure. On another hand you can use voltage to mitigate this, but you will increase temperature, yada yada. Moving along, so LN2 helps your overclock by enhancing your chip conductivity.
Moreover by reducing its temperature you will also reduce leakage currents thus reducing failure due to high leakage.

I appear to be making the case agains my argument, but wait this is the punch line. Some chips will require a minimal amount of leakage to work, this is due to their design, so if you go to low they won't work, welcome to cold bug !! So this is the overclockers world balance, where you want to go as cold as possible to maximize conductivity, but you don't want to go so low that you won't have any leakage at all.

This is why high leakage chips do well in LN2, they don't have problems of getting the leakage current too low to the point hey will malfunction or not function at all but they end up really small due to the -180?C , and they have their thermal conductivity way up top !


Hope I was able to write an understandable text

[SubZero.it]

This "formula" is completely wrong. RTFM

No, that IS the formula.
As for this generation of CPUs I'll skip again, I just suggest it for new buyers.

[kl0012]

Wtf u smoking. Ofc more power consumption = more output heat when talking about microprocessors. Or do you think that used power somehow vanishes in thin air?

http://en.wikipedia.org/wiki/Heat

In physics and chemistry, heat is energy transferred from one body to another by thermal interactions.

Basicaly it says that heat is function of thermal interface.

Its correct mostly

P=CV2f

Some notes on it from Intel

http://software.intel.com/en-us/blog...o-obvious-pt-1

http://software.intel.com/en-us/blog...obvious-pt-2-2

You can treat the notes as manual.

Although I see word "heat" there it has nothing to do with "power consumption = more heat"