Intel has solved the temperature problem in Haswell ?

[eXa]

Probably. Anyways, there is just too much variables and unknowns so far to tell if the tim issue is resolved or improved.

[sin0822]

how can an onboard VRM not add heat? You are taking more components which aren't on the CPU and putting them there.

For instance let's look at the Z68 PCH Versus the Z77 PCH. There is a TDP difference and most can confirm their PCH temps are high for Z77 compared to their Z68 boards. Z77 versus Z68 big difference is the addition of a USB 3.0 controller supporting 4 ports(and it is why the TDP is higher, i asked an intel PCH expert). The stock TDP of Z68 s 6.1W and that of the Z77 is 6.7W, that is almost a 10% increase in TDP. Every new feature for the most part adds heat, but the trick is to add more components after you take down the TDP of your CPu alone. So cut down the power consumption compared to ivy bridge, then you can add on some more features and take up the TDp a tiny bit but increase performance or add a feature.

[Hornet331]

It's an Intel CPU. They tend to run hotter than other CPUs, going back decades. They actually fried an egg a on PIII 600 and video taped it... Ivy Bridge has heat issues due to the poor FinFET design in addition to the TIM used. Intel has officially abandoned that FinFET design as a result.

[sin0822]

take a look here at more haswell OC information from IDF2013 taking place now: http://sinhardware.com/index.php/blo...l-overclocking

[zir_blazer]

Look at how ivy bridge power consumption scales with frequency alone. If haswell is similar, that 3ghz is significant.

What is wrong with that? Power consumption scales linearly with Frequency assuming Voltage remains the same, so that graph is just as expected. You may want to check this and this.

[kromosto]

What is wrong with that? Power consumption scales linearly with Frequency assuming Voltage remains the same, so that graph is just as expected. You may want to check this and this.

linear increase is expected of course but the slope is bigger then expected at least it is for me

[eXa]

Im not shure if you are directing that at me or not, thanks for the link anyways. But i am aware of that, thats why i dug up that graph, to show that you cant claim that the tim issue has been resolved based on a 3ghz sample.

Oh btw, forgot to credit Sin0822 in the first place, found it in his thread, its not my graph. So thank you Sin0822

[zir_blazer]

linear increase is expected of course but the slope is bigger then expected at least it is for me

Compared to what is bigger than expected? In one of the links I provided earlier, the guy did a ton of test with his Sandy Bridge at steps of 100 MHz to try to get accurate power consumption:

Clockspeed (GHz),Temperature (?C),Vcc (V),Power-Consumption (W)
5.0,94.0,1.487,359.0
4.9,93.0,1.488,355.0
4.8,92.0,1.488,352.0
4.7,91.0,1.488,347.0
4.6,90.0,1.488,342.0
4.5,89.0,1.489,338.0
4.4,88.0,1.489,335.0
4.3,87.0,1.489,331.0
4.2,86.0,1.489,327.0
4.1,85.0,1.489,323.0
4.0,84.0,1.490,318.0

Basically, for every 100 MHz step, Sandy Bridge consumes 4-5W more at the same Voltage *AND* temperature, while in the graph that got posted, Ivy Bridge seems to be around 3-4W per 100 MHz step with unknow temperature info. Ivy Bridge does BETTER than Sandy Bridge. However, I don't get why that Ivy Bridge graph appeared here considering that we've got nothing similar about Haswell to compare it with to know how Haswell does, and I doubt there will be any in depth analysis of Haswell power consumption scaling at least until launch.


I also don't get why people always talk about "temperature problem". It WOULD be a problem if under normal operation you reach throttling temperatures, like what happened with the infamous 3.8 GHz P4 Prescott that with its retail heatsink used to throttle most of the time. I suppose that these times Intel is intelligent enough to provide appropiate measures for that to not happen at the nominal Frequency, but after that is done, if the Processor runs at 30?C Full Load or 80, is not really important for as long that it works properly. So it makes sense that if they could reduce manufacturing cost using thermal compound instead of solder, they choose doing so on purpose, because the product still works the same while they make a bigger profit margin. So I wouldn't call it a problem, because it is intended by design.
Keep in mind that the only people that gets hurted by this are overclockers, because the lesser performance of the thermal compound vs solder when transfering heat hurts overclocking headroom. But from Intel point of view, that is not important considering how little market it represents, and that they still are the better performing option, so there is no real need to improve and their Processors will still sell.


Also, when comparing Ivy Bridge vs Haswell on power consumption and heat, I agree with what was said about the VRMs. However, there are a lot of details that I don't know if are available in some form to correctly speculate about it: First, I don't know if the VRMs were integrated on another chip like it is being done on Haswell before, to at least have an idea of its impact on another sample. Second, the VRMs dissipates heat, that in Haswell, will be obviously considered under the TDP budget (Add in the power consumption of the bigger GPU, too) that the Heatsink of it must be able to handle. But I don't know how efficient the regular VRMs on Motherboards usually are (Nor the one on Haswell itself) to know how much it actually represents. And third, even through Haswell got a slighty bigger die size than Ivy Bridge and thus with the same power dissipation, I would expect it to be slighty cooler than a comparable Ivy Bridge assuming all other being the same (TIM, Heatsink, Fan), hotspots on the die could matter a lot. Albeit I know nothing about IC desing, I suppose that the integrated VRM could be a very important hotspot (Considering how hot they usually get on both Motherboards and Video Cards), but not sure if similar or worse than the Core logic. Maybe the VRM hotspot is so bad that it becomes the most limiting factor when overclocking, even if the rest of the die is relatively cool.

[kromosto]

@zir_blazer you got me wrong i am not saying you are wrong or something and really i am away from computer tech for 3 years. just for me 4 - 5 W for 100mhz for same volt seemed to be high. i don't know other modern cpus so this may be normal but it seemed high for me. so please threat me like a noob and forgive me i am not trying to argue with you

EDIT: actually it is 5 years i am really away since 2008 just got some info on gpu's

[Manicdan]

power consumption is also based on temp, so even though you have a chart showing watts per ghz, it needs to be at a very similar temp in all cases to validate the slope.

[PatRaceTin]

I expect much on Haswell....

[sin0822]

What is wrong with that? Power consumption scales linearly with Frequency assuming Voltage remains the same, so that graph is just as expected. You may want to check this and this.

Those power consumption are at the 12v DC connector for CPU power input, i made that graph above.

[PedantOne]

Solder FTW?

It is not possible to solder 22nm and smaller node cores, forget about this for ever!

PS. many Sandy Bridge/Westmere/nehalem has TIM too, not solder!

[Kallenator]

Do you have a source on that or any information explaining why it is not feasible?

[sin0822]

Maybe the top of the die itself can't handle the heat needed for the solder process? or maybe the material can't? That would be my guess.

[STEvil]

If 22nm cannot be soldered it is only due to the top layer of the die being changed, much like Athlon Thoroughbred B had a metal layer on top that A did not. Node size does not change ability to solder, only what is done to the top layer does.

[G.Foyle]

Today's CPUs are packaged in flip-chip way - the side with transistors and interconnect layers is turned down, towards the substrate. There's nothing important on the visible side of the die. You could even sand it a bit and it'd still work.
If anything would prevent soldering, it's probably the underfill material that protects metal balls/pillars that connect the die to the substrate.

[rge]

Intel in multiple white papers developed solder tim ONLY for use with high power density, high TDP cpus (despite higher cost/lower yields from solder issues), that required solder for meeting stock cooling specs when tested under worst possible conditions. (Also there are more layers than just solder when solder is used). All other cpus still today use paste...ie cpus in 100-130W TDP range are typically soldered. CPUs in 15W typically dont have heatsinks, ie mobile, cpus in 50-70W TDP range typically have tim paste.

If IVY and Haswell were soldered, it would be intel no longer applying same logic they have been for past 15 years. The power efficiency of IVY, etc, simply put them in range where tim works fine at stock.

Load temps for IVY are low despite power density.. it is efficient at stock with low watts dissipated. It would be a bad business decision (stockholders) for intel to use solder.

Sure, when heavily overclocking the dissipated power at high power density would then make sense to use solder. But overclocking specs is not what intel uses to determine need for solder, nor has it ever been. Asking intel to set up a different manufacturing process for the less than 1% who buy K processors to overclock them, isnt exactly going to go anywhere.

I would rather the K cpus be soldered as well for maxing overclocks, but I am a realist...

[hersounds]

If intel make "K" chips for overclockers... "K" chips can be designed for overclockers and that includes SOLDER or at least best TIM bro!.

if they make an "K" chip is because that 1% is important to intel....no just overclocking. is marketing, is who have " the win"..the "faster"...come on!

[Splave]

TIM is fine! Some chips will always be better than others...

[zir_blazer]

If intel make "K" chips for overclockers... "K" chips can be designed for overclockers and that includes SOLDER or at least best TIM bro!.

if they make an "K" chip is because that 1% is important to intel....no just overclocking. is marketing, is who have " the win"..the "faster"...come on!

Why would Intel do that on the first place? They are not even "designed" for overclocking, there is no extra cost added onto R&D or bill of materials for these. The K series CPUs are just a variant of the normal ones that they can sell for 20-30 U$D higher just for giving you the Unlocked Multiplier and a better GPU (At the cost of losing VT-d, vPro and TXT, so its not even a win-win scenario because you lose a very useful virtualization feature). And it will sell to that 1% simply because it is still superior to what rival AMD has to offer. Remember that Intel is a corporation, so its basically capitalism at its finest, they aren't going to move their ass to offer you a better product if they don't have any sort of pressure to do so. Or did everyone already forget what was the driving ideology behind NetBurst architecture and how AMD hurted they market share before Conroe?

[hersounds]

Why would Intel do that on the first place? They are not even "designed" for overclocking, there is no extra cost added onto R&D or bill of materials for these. The K series CPUs are just a variant of the normal ones that they can sell for 20-30 U$D higher just for giving you the Unlocked Multiplier and a better GPU (At the cost of losing VT-d, vPro and TXT, so its not even a win-win scenario because you lose a very useful virtualization feature). And it will sell to that 1% simply because it is still superior to what rival AMD has to offer. Remember that Intel is a corporation, so its basically capitalism at its finest, they aren't going to move their ass to offer you a better product if they don't have any sort of pressure to do so. Or did everyone already forget what was the driving ideology behind NetBurst architecture and how AMD hurted they market share before Conroe?

Well bro! so explain this... why Intel do this?..."extreme edition and K skus are made for overclocking "

[zir_blazer]

Well bro! so explain this... why Intel do this?..."extreme edition and K skus are made for overclocking "

And you take what someone's marketing department says at face value? The words "designed" or "made for" are overhyped. They design thinking on that other 99% of the market, then check how they can make it more pallatable for the 1%. The K series is a product of that. They started to provide an Unlocked Multiplier as feature when they figured out they could charge more for it.
Also, just think for a second what both Intel and nVidia offers as their most high end products for the enthusiast market: Dies intended for Xeons (Sandy Bridge-E) and Teslas (GK110) that usually are sold for twice or triple the value, but they have been crippled in one way or another with disabled parts or features then sold to the enthusiasts. For many products, the enthusiast market looks like nothing more than a minor cash cow where they can sell their enterprise lines leftovers.

Chances are they profit from that one, too, they're not doing any charity with it. That price should be based on an insurance statistical model, after all, they also have accurate RMA numbers to build one.
As far that I'm aware, a CPU isn't easy to damage unless you carelessly overvolt, they're possibily the most resiliant computer part. So Intel could also make profit out of selling you extra insurance. Not everyone that pays those extra 20 U$D for that insurance will overclock and overvolt it to values where they can purporsefully kill the Processor to make use of it.

[sin0822]

you can buy the plan whenever you want from what I understood, you don't need to buy it at the counter.