Intel's Ivy Bridge Hotter Than Sandy Bridge When Overclocked

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
The company noted that it is also using “a different package thermal technology”, that “thermal technology” is known as thermal paste, which replaced a soldered heat spreader used on Sandy Bridge processors.

So now we are back to how it was with AMD's socket 754. ie. removing the shim to get better thermals...
 
It makes sense that they run hotter. Until now, decreased surface space has been proportional to decreased power so there was not much of a change from one process to the next. Now that they need to use 3d transistors, the die size has shrunk more than the power savings and we are left with chips that are a bit harder to cool. This should have been expected as soon as Intel revealed how they got to 22nm.
 
[citation][nom]SteelCity1981[/nom]Because it is. sandy bridge is on a larger die and the heat shield is solderd on and it runs cooler then ivy bridge which is a problem because normally with smaller dies comes less heat generated from the cores itself as we seen many times in the past an an example.[/citation]

Die shrinks generally weren't big changes. They were often bigger changes, but let's consider why they were often bigger changes.

Ivy has been shown to not use a whole lot less power than Sandy. However, it is a good deal smaller, especially if the IGPs are taken out of the equation. So, it's using a little less power, but it's much smaller. It's probably generating less heat, but the heat is being generated in a much smaller area. On top of that, Intel's paste is probably inferior to the solder for heat transfer. So, if Intel's next family of chips (presumably Haswell or another stepping of Ivy) improves performance per watt more than it improves performance and Intel uses a paste/solder that is better at heat transfer, then this excessive heat problem could be solved.
 
From what it sounds like, they want a shorter lifespan of these chips (in terms of performance) so they limit the OC capability via heat. They dont want chips to be as good at overclocking as SB was (the staying power of SB is huge, lets face it), but they also want increased stock performance and such... so they cut costs and limit the chips this way. Makes sense to me.
 
The reason why intel did it this way was so they can put a cap on the performance coming out of the 1155 socket. Dont let them fool you, 22nm process takes less power and less heat accordingly. To say that the process is the issue of heat is nonsense. If they did their typical soldered package on ivy bridge, the temps would be way lower.
 
Ivy bridge is made to go in ultrabooks and mac laptops, Apple has been on intel's case for years to get their chips power consumption down, and Apple buys a whole crapload of intel's production, much more than the avreage overclocker. Face it Intel is just going where the $ are, their stockholders are happy lot, including the overclockers' retirement fund managers! Ivy Bridge is designed for low power and long battery life!
 
[citation][nom]tomskent[/nom]The reason why intel did it this way was so they can put a cap on the performance coming out of the 1155 socket. Dont let them fool you, 22nm process takes less power and less heat accordingly. To say that the process is the issue of heat is nonsense. If they did their typical soldered package on ivy bridge, the temps would be way lower.[/citation]

[citation][nom]TheRabidDeer[/nom]From what it sounds like, they want a shorter lifespan of these chips (in terms of performance) so they limit the OC capability via heat. They dont want chips to be as good at overclocking as SB was (the staying power of SB is huge, lets face it), but they also want increased stock performance and such... so they cut costs and limit the chips this way. Makes sense to me.[/citation]


Those are some fairly accusatory claims there... However, they do make sense. Since paste replaced the solder, maybe we can remove the IHS and do a better job of it ourselves, for those of us who are feeling adventurous.
 
[citation][nom]jimmysmitty[/nom]Its interesting how people think it was a cost saving method when in reality the amount saved by this is probably minimal per CPU since they procude more than anyone else.I don't know why they did but even if they stuck with the solder method, there is no way for anyone to be 100% sure it would be any better since the smaller die size does add to less space to allow for heat dissipation.I read somewhere that Intel is possibly planning on moving back to the solder method and that would be a different stepping but we will have to wait and see if that actually happens and if it does the results will be clearer there.I personally think the only way to see lower temps with IB is to have more cores or more cache.[/citation]

lets put a stupid number out there... 100,000,000 units
lets say that the cost saveings was 25 cents a unit.
the lower cost isnt enough to justify a price reduction, so that means that if it cost 100$ you wont see it for 99.75.
this means that you just reduced cost without reducing price
in which case, that 100mill units just gave you another 25 million in profits

realistically, you don't have a choice if you are a high end, or over clock enthusiast right now, and sense that is a VERY small market to begin with, why should intel care?
 
So, it´s a layer of metal (heatsink) over a layer of thermal paste, over a layer of metal (shim) over a layer of thermal paste. Sounds like Inception.
 
Unfortunately, this article doesn't address the i7 3770 without overclocking at all. I'd like to see reviews on that chip w/o overclocking to see if heat is still an issue. Will that chip run a consistent hotter too? I'd like to know if the stock cooler from Intel is fine or not for the i7 3770 too.

Or, I'd like to know if Intel plans on making some changes on Ivy to address this heat issue. I'm not buying the Ivy i7 3770 if it's going to die on my after just 2 or 3 years due to heat exhaustion.
 
[citation][nom]WhateverAppleWants[/nom]Ivy bridge is made to go in ultrabooks and mac laptops, Apple has been on intel's case for years to get their chips power consumption down, and Apple buys a whole crapload of intel's production, much more than the avreage overclocker. Face it Intel is just going where the $ are, their stockholders are happy lot, including the overclockers' retirement fund managers! Ivy Bridge is designed for low power and long battery life![/citation]

Ivy is only about 7% more power efficient than Sandy (going by Tom's review of them, the 3770K averaged 3.7% faster while using 3% less power during the full benchmark), so it's not really a huge benefit right now. This could be just because Intel put the voltage settings higher so they could use more power without increasing performance. A simple one or two page review to see how well Ivy under-volts might help to clarify this a little.

Intel might have made lower power versions because Apple asked them to, but Ivy has nothing to do with that.
 
[citation][nom]becherovka[/nom]So what exactly is the purpose of the heatspreader then? It just a thermal resistance now. Need to work a good way of not using it at all. maybe a plastic shim on each side of cpu and cooler straight on top?Someone will do it without causing damage! Might need a different sort of heat paste to allow for cushioning?[/citation]

Heatsink in direct contact with the CPU while supported on the sides by shims? Remember AMD's Athlon Thunderbird and Duron processors?
 
[citation][nom]blazorthon[/nom]... Or, Graphene, diamond, or a wide host of other materials could be used to replace silicon. ...[/citation]

Latest development: silicene, the first single-atom layer of silicon, so far successfully applied to a
substrate of silver. Long way from commercial apps, but we're getting there. Step aside graphene. 😀

Ian.


 
[citation][nom]mapesdhs[/nom]Latest development: silicene, the first single-atom layer of silicon, so far successfully applied to asubstrate of silver. Long way from commercial apps, but we're getting there. Step aside graphene. Ian.[/citation]

Well, if you had quoted more of my comment, then you would have seen the part where I said that we might just have an advancement in silicon that let's it scale down to smaller process nodes than current silicon tech theoretically can instead of replacing silicon completely.

However, if silicene needs silver, it might increase the cost of manufacturing. We'll need to see if the smaller size allows it to overcome the cost of using silver (it probably does).

+1 anyway for mentioning this.
 
[citation][nom]blazorthon[/nom]Ivy is only about 7% more power efficient than Sandy (going by Tom's review of them, the 3770K averaged 3.7% faster while using 3% less power during the full benchmark), so it's not really a huge benefit right now. This could be just because Intel put the voltage settings higher so they could use more power without increasing performance. A simple one or two page review to see how well Ivy under-volts might help to clarify this a little.Intel might have made lower power versions because Apple asked them to, but Ivy has nothing to do with that.[/citation]
A very key word, and it applies to the power use too. The idle power consumption only dropped by a couple of watts, but the load power consumption dropped a lot more – close to 30 W. This results in a 16% reduction in system power at load, and the approximate CPU power (difference between load and idle power) is reduced by 31%. So the power consumption has actually dropped a lot – but not when idle (where it is probably minimal anyway).

This brings me to my next point (which is not a reply to this post): The power consumption has gone down ~30%, and the CPU (only cores) die area has approximately halved (gone down 50%, but I'll say 45% for this to be generous). Now remember a critical point: heat is the power dissipated measured in watts, temperature is something else and is measured in degrees. The temperature is proportional to the heat, and inversely proportional to the area and anything else impeding the dissipation of the heat. We have 0.7 times the heat (I'm being generous and using a 30% lower power consumption, but it is probably actually less), while having 0.55 times the area (45% less). Starting with a temperature of 70 degrees, we'll multiply it by the change in power (0.7) as the temperature is proportional to heat, and then divide the result by the change in area (0.55) as the temperature is inversely proportional to area. So we have 70 * 0.7 / 0.55 = 89 degrees – exactly where we expected it at ~20 degrees higher temperatures. If we are less generous (lower drop in power consumption, or larger drop in area), which we probably should be, we get an even higher expected temperature.
 
Oh, and I forgot: Moore's Law says nothing about transistor densities, performance, or anything like that. It predicts that the number of transistors which can inexpensively be used in an integrated circuit will double every 18 to 24 months.

http://en.wikipedia.org/wiki/Moores_law
"Moore's law is a rule of thumb in the history of computing hardware whereby the number of transistors that can be placed inexpensively on an integrated circuit doubles approximately every two years."
 
[citation][nom]JamesSneed[/nom]... Wonder if a Thermalright cooler would do better on these CPU's? I ask because the base on their coolers used to be concave(not sure if they still are) which should reduce the space between the thermal paste and the cpu plate if it bends in a bit.[/citation]

Not sure about their latest coolers with IB, but the standard TRUE and suchlike work
very nicely with SB. I'm in the process of building a 2600K setup for an artist who will
be using it for AE (Quadro 600 gfx, 90GB SSD for Windows, SAS storage for video with an
LSI3041E-R, initially 8GB DDR3/2133, ASUS P8Z68-V which cost very little). He didn't
want the highest oc, just something totally stable & long lasting. I sourced used parts
where I could, though the CPU, fans, RAM and SSD were all bought new. I bagged a used
TRUE for 15 UKP off eBay (boxed, good condition), lapped it, fitted two Coolermaster
Blade Master 120 PWM fans, it runs very nicely indeed at 4.7GHz via Offset mode (so idle
voltage & power consumption are low), needing just 1.32V for full stability. Prime95
load temps are less than 75C, noise levels excellent (Q-Fan profile working nicely). I'm
sure it could run at 5+ no problem but the intended user doesn't want to go that far
(understandable).

SB really is very good for this, so maybe the earlier poster is right, perhaps Intel is
indeed fiddling with things a bit to make IB and later CPUs not quite so brilliant for
oc'ing, lest it harm their sales elsewhere. Hard to say.

I can't be the only one talking to professional users about how these CPUs can match or
beat much more expensive fixed-speed dual-XEON systems, eg. for AE rendering the 2600K @
4.7 is 25% faster than my dual-XEON X5570 Dell T7500 (8 cores, 16 threads, 3.2GHz with
Turbo), normally a very costly system (I built it up bit by bit over 2 years). There are
newer XEONs of course, but then one could just build an oc'd 3930K setup instead and
achieve the same performance and price/performance advantages.

Who knows, it could just be a basic business decision on Intel's part: if they make
products that are too good, too long lasting, and it could harm their tick/tock cycle
wrt demand (depends on the market I suppose). I've been talking to an admin at one major
movie company about the possible cost advantages of using oc'd quality consumer builds
as the basis for lesser scale renderfarms (it's not as if the process is time consuming
anymore - it took me just 10 seconds to setup 4.5GHz stable, almost too easy with SB);
looks like the only thing holding it back is the lack of will at a senior level to try
out the idea.

Ian.

 
[citation][nom]PreferLinux[/nom]Oh, and I forgot: Moore's Law says nothing about transistor densities, performance, or anything like that. It predicts that the number of transistors which can inexpensively be used in an integrated circuit will double every 18 to 24 months.http://en.wikipedia.org/wiki/Moores_law"Moore's law is a rule of thumb in the history of computing hardware whereby the number of transistors that can be placed inexpensively on an integrated circuit doubles approximately every two years."[/citation]

Saying that the transistors able to be put in a specific area doubles every x months (it started off at 12 months or so and slowed down over time) is like saying the transistor density for the inexpensive chips doubles every x months for chips. The two are synonymous.
 
[citation][nom]blazorthon[/nom]Well, if you had quoted more of my comment, then you would have seen the part where I said that we might just have an advancement in silicon that let's it scale down to smaller process nodes than current silicon tech theoretically can instead of replacing silicon completely. ...[/citation]

Indeed, there are all sorts of possible new approaches. IMO material science is the most important
science of all for the modern tech age, yet the least publicised in schools, etc. Everything else is
built upon it.


[citation][nom]blazorthon[/nom]... However, if silicene needs silver, it might increase the cost of manufacturing. ...[/citation]

Ah, I should have clarified, the silver base was just used to show it was possible to lay a single atom
silicon layer at all. The team involved are now going to see if they can achieve the same thing with a
base made from an insulating substance. See:

http://www.newscientist.com/article/mg21428625.400-move-over-graphene-silicene-is-the-new-star-material.html

It's early days yet.


[citation][nom]blazorthon[/nom]... +1 anyway for mentioning this.[/citation]

Thanks!! Hmm, someone gave you a minus, dunno why, so I negated it.

Ian.


 
It's quite obvious that thermal paste will always be more resistant to heat transfer than soldered aluminum. Well, whatever is cheaper for Intel.

On the plus side, one might get a lower 1-2 C if they removed the glued heat spreader and directly attached the heatsink/waterblock to the die cover. Just don't apply too much pressure or Ivy Bridge will become Ivy Sand.
 
If you do not think Ivy Bridge was designed with ulterbooks in mind then why does it perform so well when the power is reduced below refrence voltage!
 
[citation][nom]blazorthon[/nom]Saying that the transistors able to be put in a specific area doubles every x months (it started off at 12 months or so and slowed down over time) is like saying the transistor density for the inexpensive chips doubles every x months for chips. The two are synonymous.[/citation]
But the thing is that it has nothing to do with area. It is saying that the number of transistors that can be put in an IC (of any size) inexpensively (enough that the price has decreased due to quantity but not enough that there are too many defects) will double every 24 months (which is what was originally stated – not 12 months).
 
[citation][nom]becherovka[/nom]So what exactly is the purpose of the heatspreader then? It just a thermal resistance now.[/citation]
In the good ol' days, CPUs were mounted on the pins-side of a ceramic substrate (386/486/Pentium/PPro) and the heatsink was simply applied to the ceramic substrate.

Then came the Pentium 2 and AMD counterpart with power going too high for through-substrate heat dissipation and caches were starting to run too fast with too many pins to go through processor sockets so both AMD and Intel switched to processor cartridges with factory-mounted HSF with more direct contact with flip-chip CPU dies.

Once they became able to integrate cache into the CPU die (Coppermine and AMD equivalent), they decided to go back to the cheaper socket interface with naked flip-chip CPU dies on organic substrate for heat-transfer efficiency and many people cracked/chipped their CPUs' dies through mishandling or accidents. CPUs also tended to burn out almost immediately if PCs were powered on with HSFs improperly attached or missing.

Next generation, we got the introduction of the heat-spreader as we know it which plays both a mechanical protection and thermal protection role: it mechanically protects the core against mild physical abuse and disperses heat from hot-spots to give thermal protection enough time to react. Without heat-spreader, hot-spots can instantaneously damage the core if there is bad/no heatsink contact such as when HSFs fall off during shipping of pre-assembled PCs.

I doubt many people would be willing to sacrifice their metal "safety caps" for a few hundred MHz higher overclocks. If you look at this the other way, Intel using thermal paste now allows the most extreme overclockers to actually remove the cap to push they luck even further than ever before.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom