Ivy Bridge Delidded, showing traditional TIM

Ivy Bridge Temperatures - It's Gettin' Hot in Here

Add Your Comments

Why is Ivy Bridge so hot? Ask that question in any forum currently, and you are likely to receive one of two different popular (but not entirely correct) answers that everyone has been parroting:

  1. “Power density is greater on Ivy Bridge than Sandy Bridge”
  2. “Intel has problems with tri-gate/22nm”

The first answer is correct, but wrong at the same time – power density is greater, but it isn’t what is causing temperatures to be as much as 20 °C higher on Ivy Bridge compared to Sandy Bridge when overclocked. The second answer is jumping to conclusions without sufficient evidence. If you aren’t in the loop, there’s evidence of a considerable temperature difference nearly everywhere you look – we confirmed it by mirroring settings in our Ivy Bridge review, and we have read similar reports in solid testing at Anandtech as well as from other sites.

So why is Ivy Bridge hot?

Intel is using TIM paste between the Integrated Heat Spreader (IHS) and the CPU die on Ivy Bridge chips, instead of fluxless solder.

Ivy Bridge Delidded, showing traditional TIM

Ivy Bridge Delidded, showing traditional TIM (Image courtesy of pt1t.eu)

 

How does TIM paste generally compare with fluxless solder for conducting heat? Heat conductivity can be measured in watts per meter Kelvin. To be technically exact, we would need to know exactly what Intel is using for TIM paste/solder. When I went to Intel and asked, their polite answer may not surprise you – “Secret sauce”!  Given that, we can use some rough approximations. A solder attach could have a heat conductivity in the range of 80 W/mK. A TIM paste could have a heat conductivity in the range of 5 W/mK. That’s your problem right there! Note that these values are not exact, as we don’t know the exact heat conductivity of Intel’s “Secret sauce”. However, these are values representative of solder or TIM paste, and there is a giant gap between how TIM paste and solder perform in regards to conducting heat. They are in different leagues.

Demonstrating the Problem

Most importantly here, if Intel is using TIM paste between the IHS and CPU die, the IHS effectively becomes a heat barrier rather than a heat spreader. Here is a rough diagram of the current heat transfer on Ivy Bridge:

  • CPU Die -> 5 W/mK  TIM -> IHS -> 5 W/mK  TIM -> Heatsink

It would be far more beneficial for temperatures to take a more direct route such as:

  • CPU Die -> 5 W/mK TIM -> Heatsink

Extra heat interfaces are a bad thing, especially when they have relatively low thermal conductivity. On a fundamental level, it doesn’t make much sense to do things this way from the perspective of optimal cooling. However, it could make sense from a die-protection standpoint.

In contrast, a fluxless solder attach like that described in Intel patents was invented for the specific purpose of quickly and effectively radiating heat away from the CPU die. In this situation with a solder that can conduct heat in the range of 80 W/mK and in light of tighter and tighter power densities as Intel continues to shrink its processor die, you can start to  see on a fundamental level how quickly getting the heat from a very small area to a slightly larger area may be helped by the design of a soldered IHS. This still leaves the problem of a 5 W/mK TIM paste interface between IHS and heatsink, but before you get there you have a high conductivity solder attach between die and IHS that radiates the die heat to a larger area.

Ivy Bridge Power Density

Power density likely became a popular answer because Intel has referenced the challenges it presents with process shrinks, and it just makes sense on basic level. Very hot die, smaller area to conduct heat away from. Blaming power density for the heat issues is easy! However, Ivy Bridge has approximately 75% the die size of Sandy Bridge, which is a big difference certainly, but not enough to explain the stark contrast in temperatures obtained by our peers across review sites and the forums. Where Sandy Bridge would often be around the 60 °C range at a 4.5 GHz overclock, Ivy Bridge has been tested to be in the 80-90 °C range.  How can we blame power density for a difference that large? That dog just doesn’t hunt!

In light of this contrast, we can gain further insight as well from what history has taught us. If you’ve been paying attention, we saw similar issues between the E6XXX and E4XXX processor lines. The E6XXX used a solder attach under the IHS and were far easier to keep cool. The E4XXX used a TIM paste under the IHS and ran hot! Those aren’t the only examples, and I’m certain enthusiasts in the community with better memories than myself can lend further supporting evidence from our past experience. Given hindsight, it is hard to explain why Intel would make a return to TIM paste for Ivy Bridge.

Bottom Line

So based on what evidence we could find from our own investigation, as well as what experience has taught us, Ivy Bridge is running hot when overclocked because of TIM paste between the IHS compared to solder attach used on Sandy Bridge. Why Intel made this choice we aren’t yet sure. We also aren’t sure if they will continue using TIM paste on the Ivy Bridge line, or if this will only be seen on the Engineering Samples like the units sent out for review.  However, we’ve put word out again to Intel and are waiting to hear back if they have any further insight or comment to offer. If nothing else, we can hope their reply will again be in good humor… “Secret Sauce” did give us a laugh!

I.M.O.G.

Leave a Reply

Your email address will not be published. Required fields are marked *

Discussion
  1. I'm glad somebody took the time to explain all that. I'm planning on buying a 3770K around December, so I'll definitely be keeping up with Intel's statements. They've already publicly stated that Ivy Bridge runs hotter, though, so I can't imagine them switching to a better heat conductor after already publicly denouncing their CPU's ability to stay cool. Maybe TIM is cheaper, and they've decided that with AMD out of the picture they can do whatever the hell they want.
    I didn't want to wander too far into the weeds within the article itself, but I want to mention as well about delidding. Many people know it can be dangerous and easy to kill a chip that way. However people often forget a challenge with making contact with the die - if you delid, you may need to modify the socket/socket retention clip to ensure the base of your cooler can contact the die, as without the IHS the chip will ride lower on the motherboard. Just a note for anyone who may follow up with further testing on this.
    It's also important to note that those who air cool their CPU with direct-contact heatpipe coolers do NOT want to delid their CPU. You'll have heatpipes that don't touch the heat-producing die surface and be worse off than when you started.
    I.M.O.G.
    I didn't want to wander too far into the weeds within the article itself, but I want to mention as well about delidding. Many people know it can be dangerous and easy to kill a chip that way. However people often forget a challenge with making contact with the die - if you delid, you may need to modify the socket/socket retention clip to ensure the base of your cooler can contact the die, as without the IHS the chip will ride lower on the motherboard. Just a note for anyone who may follow up with further testing on this.
    Also, you want to make sure you do not crush the die either. :thup:
    Great and informative article IMOG!! :clap:
    Sort of similar deal with most recent waterblocks as well Hokie, they are designed to cool a heat spreader, not a focused central die - older blocks like the Storm were designed to cool bare die.
    As for crushing the die, I probably won't be delidding myself - my chip will only see LN2 usage, and with remounting the F1EE on a weekly basis, I'd be bound to crack or crush the die before long. In that case, I'll take the IHS for protection alone - the LN2 will do its job well enough. A lot easier to hurt a chip without the IHS on.
    could you take the lid off, clean up tim, put some little chunks of solder were the tim was set lid back on and heat it with a heatgun or mini torch until it settles back down :D could work? :shrug:
    very nice find, and also very informative.
    one question I would have is if sandy bridge has been confirmed to have fluxless solder and not a similar paste. I would guess it is the solder method, as it would explain the huge discrepancy on temps between 2 fairly similar cpu's. I just would love to see a delidded sandy bridge cpu as well.
    What are the differences in waterblocks designed for direct die cooling vs IHS cooling? I would expect completely flat base for direct die, but what else?
    The whole direct die cooling era was before my time :D
    MattNo5ss
    What are the differences in waterblocks designed for direct die cooling vs IHS cooling? I would expect completely flat base for direct die, but what else?
    The whole direct die cooling era was before my time :D

    There was something of a pinnacle in water block design, just before the time when the IHS first became popular. Once the IHS came into the picture, blocks designed for use without an IHS were less effective with an IHS, and new blocks adjusted accordingly.
    With waterblocks designed to cool a die without an IHS, microchannels and impingement (don't hear that too often anymore!) were major design features. Impingement over the central area of the die was important in that it greatly increased the interaction between the water and surface area directly above the die. There was more focus on cooling that small area primarily, than the current focus on cooling the larger IHS area.
    There are other old timers who are much more well versed than myself within our community who could speak in more detail and with more knowledge. Nikhsub knows, cathar knows, billa knows... Others too, but those are the guys off the top of my head. rge is still very active around here and he could probably explain in more detail than me as well.
    Didnt SB use the same TIM on the IHS and it has no problem.
    This analogy has no facts to prove this theory either a test of removing IHS and have direct contact needs to be done to confirm.
    If this theory was true then water cooling would not help. The heat would build up and not make high enough dissipation for water cooling to be affective.
    A heat barrier is like not locking down your heat sink then the temp is high and nothing will cool it.:)
    I would be very surprised if they would switch this "secret sauce" for solder on later chips... but who knows, would be something interesting to wait for then!
    Nice editorial Matt! Glad to see the discussion from the Ivy review got made into an informative frontpage article.
    Hopefully the TIM is only on the review samples.
    bmwbaxter

    Hopefully the TIM is only on the review samples.

    What's next? Lamborghini sending Road and Track Lambos with tires from Wal-Mart? Oh, wait, they have competition.
    very nice once again Matt. I suspected that SB was soldered, proof is in the pudding.
    rather dissapointing that Intel would back off a few nickels savings for the heat terror that those jumping to ivy will be faced with. rather sad day for me. think I will wait it out and see if they change thier minds on a new stepping.
    Hoping this thread exposes their choice enough to make them rethink the process
    Thanks Jonnie. I'm not totally convinced that TIM paste will be a permanent feature really... One might actually suspect that retail units will be soldered, and what we've found and explained in this article is more meaningful for the preliminary results we're seeing in most of the reviews and forums about Ivy getting hot when overclocked.
    There are some exceptions, despite most people raising questions about the heat when overclocked, like the computershopper overclocking results:
    http://computershopper.com/components/reviews/intel-core-i7-3770k/(page)/4#review-body
    They didn't perceive the temperature issues reported in most articles with their sample. Their test was rather quick at 10 minutes then a temp reading, but I suppose it could be an indication that not all these chips are hot. Anxious also to see if Intel can add better insight than what we have to work off of currently.
    I'm glad I'm still not testing water blocks... I think a Storm would be ideal on a bare die processor for H2O cooling... you do NOT want a bowed block on a bare die - at all. We are back to flat as can be again. No one is making blocks for bare dies at all. Intel really needs to be consistent, either solder the lids like you started doing a few years back or just get rid of the IHS altogether (doubt that will ever happen).