My thoughts on ivy bridge 22nm heat problem

[nyker96]

My thoughts on ivy bridge 22nm heat problem

first of all, intel has gone through many iterations of process shrinks, in all previous generation the die shrink makes transistor concentrate more tightly, consequently increasing heat/area. However, the smaller process also means lower voltage is needed, this seem to offset the increased heat/area problem. So it's something that had occurred before and was never a problem.

However, this time, the 22nm process seem to be so sensitive to voltage increase that it results in incredible increase of heat/area out pacing the fact that less voltage is needed by the new process to drive the chip.

Since TSMCS/GF all sell different processes for different applications, some cater to mobile applications that perform well in low voltage situations, others are designed for high voltage applications. It's my theory that the intel 22nm was primarily optimized for mobile/low voltage applications. So the process becomes atrociously bad when you turn up the voltage.

If anything i think intel is starting to not put as much care on high performance markets since AMD pretty much offers no competition there. Intel is now just concentrating on low voltage mobile markets that's dominated by ARM etc. It makes sense for them to do this business-wise.

What do you guys think?

 

[VirtualLarry]

I tthink that you might be on to something there. It makes sense to me.

 

[Ventanni]

It really only gets warm when you crank up the voltage and up the clockspeed. Mine only peaks at 48C during gaming so far with a CM 212. At stock though, it's fine. But yeah, I think it was anand that said it's such a small die cause Intel probably didn't think it had to improve the speed much over sandy bridge.

 

[Tempered81]

It scales higher than 2700K's on LN2 because of the higher multiplier limit. The cheap thermal paste they used on the high-end 3770k helps to keep the average overclocker at a reasonable 4.5-4.8ghz. Not saying 3d finfet xtors doesn't impact it, because they do as well as the smaller cores and small die, but paste vs. iridium solder also contributes. That paste is the same paste under the IHSs of all the previous generation's low-end products like e4400s and celerons and pentium D's. The high end parts (QX6850, Q9450, i7-920, i5-2500k) all used high quality solder. The solder transfers heat 20-30 times better than intel's paste. In that recent Chinese article where they replaced the stock intel paste with some aftermarket tim, the thermals got closer to Sandy thermals, like 20C cooler on overclocks. The decision to use cheap paste saves them a buck and results in more usable cpus made, and keeps SBE appealing. If 3770k's could do 5.4 out of the box, who would buy the high-end? Making sure they do 4.5-4.8 keeps value in a 4.9 6-core and keeps their pricing structure in tact.

A new revision, the move to 6-core and iridium solder could make IBE or a new stepping of IB a 5g+ chip for the average overclocker.

 

[cytg111]

My thoughts on ivy bridge 22nm heat problem

first of all, intel has gone through many iterations of process shrinks, in all previous generation the die shrink makes transistor concentrate more tightly, consequently increasing heat/area. However, the smaller process also means lower voltage is needed, this seem to offset the increased heat/area problem. So it's something that had occurred before and was never a problem.

However, this time, the 22nm process seem to be so sensitive to voltage increase that it results in incredible increase of heat/area out pacing the fact that less voltage is needed by the new process to drive the chip.

Since TSMCS/GF all sell different processes for different applications, some cater to mobile applications that perform well in low voltage situations, others are designed for high voltage applications. It's my theory that the intel 22nm was primarily optimized for mobile/low voltage applications. So the process becomes atrociously bad when you turn up the voltage.

If anything i think intel is starting to not put as much care on high performance markets since AMD pretty much offers no competition there. Intel is now just concentrating on low voltage mobile markets that's dominated by ARM etc. It makes sense for them to do this business-wise.

What do you guys think?

I think it is a possibility in the cloud of infinite possibilites (a whole lot atleast).
Many things can be at the table here. Look down memory road, at every tick tock toe, process shrink, etc seems to come with a somewhat large "risc" factor .. alot of shit can go wrong, and i think thus became intels tick tock strategy, to manage those riscs..
I see Ivy's temps within those margins, Intel will get a chance to get the node right before rolling out the tock (or tick,as it may be)

 

[DigitalWolf]

If anything i think intel is starting to not put as much care on high performance markets since AMD pretty much offers no competition there. Intel is now just concentrating on low voltage mobile markets that's dominated by ARM etc. It makes sense for them to do this business-wise.

What do you guys think?

After reading this I guess my thoughts are:


If the lack of competition from AMD in the high performance market was driving Intel... we would not be buying products like IB for the prices we are.


Its also odd (to me) that anyone would have an expectation that the product should perform better out of spec. Now in a pure overclocking conversation I can see this type of thing being discussed. I cannot see it being much of a topic when it comes to the design of an entire chip line... that the majority of sales will very likely be run in spec (thus no heat problem).


One of the ever present trends with electronics is reducing size. Part of the issue with that is surface area and how it relates to thermals. The real "problem" here is that thermal solutions will need to change as surface area continues to shrink.

 

[nyker96]

It really only gets warm when you crank up the voltage and up the clockspeed. Mine only peaks at 48C during gaming so far with a CM 212. At stock though, it's fine. But yeah, I think it was anand that said it's such a small die cause Intel probably didn't think it had to improve the speed much over sandy bridge.

that is exactly what I am saying, at lower stock volts, or even underclocking, this thing is unbeatable. but once you crank up the voltages, it becomes like another animal.

After reading this I guess my thoughts are:


If the lack of competition from AMD in the high performance market was driving Intel... we would not be buying products like IB for the prices we are.


Its also odd (to me) that anyone would have an expectation that the product should perform better out of spec. Now in a pure overclocking conversation I can see this type of thing being discussed. I cannot see it being much of a topic when it comes to the design of an entire chip line... that the majority of sales will very likely be run in spec (thus no heat problem).


One of the ever present trends with electronics is reducing size. Part of the issue with that is surface area and how it relates to thermals. The real "problem" here is that thermal solutions will need to change as surface area continues to shrink.

I personally think the lack of competition at high end is driving Intel's decisions to optimize for low voltage market, look at haswell, it's another attempt at power saving design. Of course that's my opinion from the info I saw.

But I disagree w/ not having an OC expectation especially comes to the K-series. The regular chips they sell, no I don't suppose I expect anything from them.

of course you could be right that we may just have reached some type of turning point, that once you pass 22nm the heat/area becomes unmanageable. If this is indeed the case and wasn't an design issue, then we are screwed. From here on all the heatsinks will be insufficient to remove heat quick enough and we will probably all have to do custom water loop to OC from now on.

 

[KentState]

that is exactly what I am saying, at lower stock volts, or even underclocking, this thing is unbeatable. but once you crank up the voltages, it becomes like another animal.



I personally think the lack of competition at high end is driving Intel's decisions to optimize for low voltage market, look at haswell, it's another attempt at power saving design. Of course that's my opinion from the info I saw.

But I disagree w/ not having an OC expectation especially comes to the K-series. The regular chips they sell, no I don't suppose I expect anything from them.

of course you could be right that we may just have reached some type of turning point, that once you pass 22nm the heat/area becomes unmanageable. If this is indeed the case and wasn't an design issue, then we are screwed. From here on all the heatsinks will be insufficient to remove heat quick enough and we will probably all have to do custom water loop to OC from now on.

As Intel stated in their own marketing material, consumers are moving to mobile devices away from desktops and they are positioning themselves properly in the market. If 75% of chips sold are for mobile devices, then energy efficiency is more of a concern.