Report: Delidded Ivy-E processor has solder under its heatspreader

Sandy Bridge processors have fluxless solder between the CPU die and integrated heat spreader. Intel swapped the solder for thermal paste in Ivy Bridge, a change that was in part blamed for the chip’s high operating temperatures when overclocked. The latest Haswell CPUs use a similar thermal interface material to Ivy-based chips. They’re even toastier when pushed, and overclockers not impressed. Clock-boosting junkies may get a reprieve when Ivy Bridge-E arrives later this year, though.

A user in the Coolator forums has popped the lid on a purported Core i7-4960X, revealing old-school solder rather than the TIM used on Intel’s recent desktop CPUs. This isn’t conclusive evidence, but I wouldn’t be surprised if the report were accurate. Like Sandy Bridge-E, Ivy-E is basically a server-grade Xeon rebranded for duty in high-end desktops. It doesn’t make sense for Intel to skimp on the thermal performance of a product designed for toasty server racks. The higher margins commanded by Xeon-based parts should cover any additional cost associated with the solder, as well.

With Intel’s consumer CPU silicon increasingly focused on mobile and all-in-one systems, the Ivy-E may provide salvation for hardcore enthusiasts with a thirst for higher frequencies. The rumored specs suggest that the top six-core, 12-thread model will be clocked at 3.6GHz with a 4GHz Turbo peak. That’ll probably cost a grand or more. However, the Sandy Bridge-E lineup includes a quad-core part at $300 and a six-core one for $570. Ivy-E may have similarly priced alternatives.

Thanks to Hardcorewear for the tip.

Comments closed
    • swaaye
    • 8 years ago

    I remember my Opteron 165 had paste. All old consumer CPUs were paste until I think Core 2 or some P4. I recall first seeing guides for Core 2 soldered IHS removal so maybe that was the first.

    • chuckula
    • 8 years ago

    OK, the results below are a quick & dirty analysis of the die size seen in that chip. Short answer: That die is approximately 267 mm^2, which is not all that big for this type of chip. It could mean we are really looking at a 4-core SB-E part, or Intel has substantially shrunk the die for IB-E.

    I used the top edge of the LGA-2011 PCB, which includes notches, and is known to be 45 mm in length (http://www.intel.com/content/dam/doc/guide/core-i7-lga-2011-guide.pdf). I then drew some overlay lines in inkscape and got these pixel sizes (your units may vary, but if the ratios are about right then using the known 45 mm length of the PCB should give you a similar result).

    1. Width of the PCB in pixels (width measured along the clear top PCB edge including notches): 815px
    2. Width of the Die in pixels (parallel to PCB width): 309px
    3. Length of the Die in pixels (perpendicular to PCB width): 283px
    4. Width of die in millimeters: (309px / 815px) * 45 mm ~= 17.06 mm
    5. Length of die in millimeters using mm width as a reference: (283px / 309px) * 17.06mm ~= 15.63 mm
    6. Total die area: 17.06 mm * 15.63 mm ~= 267 mm^2 (Edit: after a second more accurate crack I got the number up to 271 mm^2.. so in that neighborhood)

    Edit: Interesting data point: From the Anandtech review of the I7 3820, the die size for the 4-core SB-E parts is 294mm2. That puts a 4-core SB-E part at the high end of a margin of error for my estimate…. So the photos could be a fake of a delidded 3820 that we all known has a soldered IHS, or the 22 nm process has a pretty good die shrink for IB-E parts.

    Edit 2: Die size for an 8 core SB-E: 435 mm^2. So this die is roughly 61% of the size of the 8 core SB-E die, which is plausible for an 8 core IB-E after a die shrink since you never get a perfect 50% size reduction with a straight up shrink. Very interestly.

      • Waco
      • 8 years ago

      Someone correct me if I’m wrong…but the 4-core SB-E parts are the same size as the 6-core and 8-core variants. Intel just lists a smaller die size because the “active area” of the die is smaller due to the reduced core/cache counts.

        • chuckula
        • 8 years ago

        [quote<]Someone correct me if I'm wrong...but the 4-core SB-E parts are the same size as the 6-core and 8-core variants.[/quote<] Corrected 😉 The 6 core SB-E parts are identical to the 8 core parts with 2 cores & some cache deactivated. The 4-core parts are using a different die however. See Anand's review of the 3820: [url<]http://www.anandtech.com/show/5276/intel-core-i7-3820-review-285-quadcore-sandy-bridge-e[/url<]

      • madmilk
      • 8 years ago

      Do we know for sure that IB-E is using a harvested 8 core die again (so 12, 8 and 4 cores)? I could see Intel only doing 12 and 6. A 6 core die would fit very well with the die size measurement.

      That would also explain the lack of a 8 core IB-E, since Intel wouldn’t want to disable 4 cores on a 12 core chip.

    • LukeCWM
    • 8 years ago

    [quote<] has popped the lid on a purported Core i7-4960X[/quote<] There's the "purported"! I found the Tech Report's equivalent of Waldo today!

    • kamikaziechameleon
    • 8 years ago

    I think this is a good sign of things to come. Good to see we can expect better temps weather you OC or not this will mean quitter computers for those who don’t OC.

    • DeadOfKnight
    • 8 years ago

    I don’t know why anyone would drop this money on an Ivy-E 6-core when it’s already been announced that Haswell-E will sport 8 cores. Just seems counter-intuitive to me unless you can’t stand your Bloomfield CPU any longer.

      • madmilk
      • 8 years ago

      And Broadwell-E will have 10 cores. Of course it comes down to “I need a new CPU now”.

        • DeadOfKnight
        • 8 years ago

        Right, but what kind of performance difference will Ivy-E even have compared to what’s already available? Haswell-E 8-core will deliver a significant improvement that actually makes sense to wait for. If there’s an ideal upgrade cadence for the enthusiast line, it should probably be matched up with extra cores. Broadwell-E might have 10 cores, but I’d be surprised if they didn’t disable some of them like they did with SB-E.

    • Voldenuit
    • 8 years ago

    [quote<]Intel swapped the solder for thermal paste in Ivy Bridge, a change that was in part blamed for the chip's high operating frequencies when overclocked.[/quote<] Geoff, you might want to edit this sentence, because right now it's saying that raising your clockspeed... raises your clockspeed? :p

      • NeelyCam
      • 8 years ago

      But… that’s perfectly factual – no need to change.

        • chuckula
        • 8 years ago

        It’s like that song… the more I drink, the more I drink.. only overclocked. Or just drinking, who cares!

      • Laykun
      • 8 years ago

      It makes it sound like an Onion article 🙂

    • xeridea
    • 8 years ago

    Sweet, so to get proper thermal dissipation, you only need to spend $600-1000+, or you could just get a $30 CPU. It is absolutely insane to think that a $350 chip has $0.05 cents worth of thermal paste, and a $30 CPU has proper solder.

    Its not just for OC, having crappy chip design requires a larger cooler, or more noise for reasonable temps under load.

    • SoM
    • 8 years ago

    i like post #8 on that site, look at the dudes speedtest *drools*

    • Deanjo
    • 8 years ago

    Think this is a bit overblown. For the vast majority of CPUs are not overclocked. Intel’s thermal paste does a sufficient enough if a job to serve that market just fine. Sure it will not satisfy the extreme fringe OC market but they always seem to think that they can do better anyways with their lapping, delidding, etc.

      • mesyn191
      • 8 years ago

      But they’re selling CPU’s for OC’ing with the shitty TIM job….

      • willmore
      • 8 years ago

      Then, would it not be reasonable to expect the K series parts to come with soldered IHS since they *are* sold very specifically for overclocking?

        • Deanjo
        • 8 years ago

        Not necessarily, you can still overclock with the TIM, you are still capable going past spec of the Non-K versions. They are offering a product that can be overclocked but not to extreme levels. They do have a line that does cater to those extremists. Hence the “Extreme” line.

          • tootercomputer
          • 8 years ago

          You sound like an apologist for Intel. My 2500K SB OCed to 4.5GHz with so little effort and I was able to keep the temps extremely reasonable with only air. For that chip, 4.5GHz was not that “extreme” of an OC. In contrast, K versions of IB and now Haswell, for an equivalent OC, run extremely hot and that complicates the entire OC process. So Intel should be ashamed for not providing a better thermal solution for a chip that is designed for overclocking.

            • Krogoth
            • 8 years ago

            Do you realize that most of the problem is from 22nm process? It doesn’t take well to overvolting. The delidded and direct contact Haswell and IB still get toasty fast when you crank up the volts.

            • NeelyCam
            • 8 years ago

            [quote<]Do you realize that most of the problem is from 22nm process?[/quote<] That in itself is not the source of the problem. Increasing voltage speeds up the circuit, but so does switching to new, faster transistors. [i<]All else being equal[/i<], the 'next-gen' circuits don't need as much voltage to run as fast as the 'prev-gen' circuits. Just look at those Intel plots where they show that, compared to 32nm, 22nm is 30% (?) faster at same power or 50% (?) lower power at the same speed. All CPUs start consuming power when you increase the voltage to go faster (you've seen what happened to the FX-9000 series). You might think that 1.5V should be doable with 22nm because some 45nm AMD CPU at was running at 1.5V. That's wrong - 1.5V is a [b<]50% increase[/b<] from the standard 1V or so. That would be like running that nominally 1.5V CPU at 2.25V. And keep in mind that temperature and power consumption has a clear, direct relationship. Let's say we up the voltage and overclock the chip to the point where it's consuming 150W of power. Then, cool that chip with a cooling solution that has a thermal resistance of 0.1K/W. With +20C ambient temperature, you get the chip running happily at +35C. But try that with a 0.5K/W cooling system... and you'll get a very toasty +95C. Note that thermal resistors can be put in "series"; e.g., there is one "thermal resistor" between the chip and the IHS (i.e., TIM and those air bubbles), and then there's another one between the IHS and the 'ambient room temperature' (your copper heatsink/waterblock, fans, case cooling etc. all play a role in determining this resistor value). If the thermal resistance between the chip and the IHS is 0.4K/W, it doesn't matter how great your heatsink and your array of bright blingbling fans are - you're still screwed - you can't bring that full thermal resistance from the chip to ambient any lower than 0.4K/W. This is not a "22nm problem". This is a cooling problem.

            • Krogoth
            • 8 years ago

            You need to crank up the volts in order to get a stable, aggressive overclock. The problem is that IB/Haswells needs volts to keep it stable when you go beyond ~4.4-4.5Ghz. 22nm process roasts up quickly when you dump volts. The IHS/paste issue just makes it worse.

            IMO, overclockers on the Intel side got spoiled rotten by how easy it was to push post-Conroe chips. IB and Haswell broken this convention and they are frustrated by this. They quickly point to the IHS/paste problem, but the die-crowd are still disappointed that IB/Haswell don’t go beyond what SB can do before hitting a thermal wall with exotic cooling.

            • NeelyCam
            • 8 years ago

            The “high-volts for high-freqs” is not a “22nm problem” – it’s an IB/HW problem. There is no fundamental, semiconductor process physics related reason why it would be a “22nm problem”

            And “high-power -> high-temps” is not a “22nm problem” – it’s a cooling problem.

            I agree with the rest of your post, but just wanted to point out that “22nm” has pretty much nothing to do with any of this.

            • willmore
            • 8 years ago

            That’s simply not true.

            Leakage current increases with decreasing gate geometry.
            Leakage current increases with increasing standoff voltage.

            Power is the product of current and voltage. So, as process geometry decreases and volage is raised, power usage will increase as a third order exponential.

            In plain language–as process geometry decreases, the range of efficient voltages (and the resulting operating frequencies) will narrow. The low end will be eaten up by the static leakage current which cannot be offset by decreased drive voltage at lower frequencies. The high end will be limited by leakage current increase due the higer voltage needed to acheive higher frequencies.

            And, it’s not just the people here, there is a general feeling in the industry that Intel’s 22nm FinFET didn’t deliver:
            [url<]http://www.goldstandardsimulations.com/index.php/news/blog_search/simulation-analysis-of-the-intel-22nm-finfet/[/url<]

            • NeelyCam
            • 8 years ago

            [quote<]That's simply not true.[/quote<] Which part? I had two major statements; First, "The "high-volts for high-freqs" is not a "22nm problem" - it's an IB/HW problem." This has nothing to do with what high voltage does to leakage or power consumption - just that high voltage is needed to run these particular chips faster. They could've been designed in such a way that they could be running faster. Just look at AMD chips - they run with much "faster" (frequency) than Intel's 32nm or 22nm chips. Is that because of the process? No - that's because of the architecture. The second one: "And "high-power -> high-temps" is not a "22nm problem" - it's a cooling problem." I think we agree on that one. Right? Note that I agree that high voltage increases power consumption - both leakage and active. Decreasing gate geometry can increase leakage, but on the other hand, changing the transistor architecture can reduce leakage in spite of decreased gate geometry. For instance, high-K gate dielectric reduces gate leakage a lot. FinFET/TriGate structure improves channel control, also reducing leakage - this is the main benefit of a FinFET vs. planar. [quote<] The high end will be limited by leakage current increase due the higer voltage needed to acheive higher frequencies.[/quote<] In that situation it's not the leakage that dominates, but active current (the CV^2 one). Thanks for the link - it was pretty interesting. I have no idea how you got "there is a general feeling in the industry that Intel's 22nm FinFET didn't deliver" from that link, though. There was only one guy named "James" (I think I've seen him on S|A forums) complaining - or, rather, trolling - that IB didn't idle well and Intel cherrypicks benchmarks and Intel marketing is lying. I'm not sure that constitutes a "general feeling in the industry". The article itself said that Intel's trapezoidal FinFET short channel effects were worse than in a rectangular FinFET, but no comparison was done to a planar transistor.. I'm not sure that is enough to talk about a "general feeling in the industry". On the other hand, the general feeling in the industry right now seems to be that Haswell has significantly improved battery life in MBAs and other ultrabooks. Should we then conclude that the general feeling is that 22nm has definitely delivered? I would rather say Haswell has delivered, and wait to see what Silvermont looks like... because that's where the promises of high efficiency are really supposed to be realized.

            • Deanjo
            • 8 years ago

            Not apologizing for intel at all, simply putting in perspective who 99 percent of their customers are. The “K” series sells because of the possibility of overclocking which is a powerful marketing tool to up sell and I am willing to bet the vast majority never get clocked any higher then what the factory preset is. For that miniscule percentile that do want extreme overclocking they do have a line for that but be prepared to also pay for it.

      • Krogoth
      • 8 years ago

      Pretty much.

      The whole thermal paste thing is overblown. It is just bean counters at Intel trying to squeeze a few cents per unit.

      • Aliasundercover
      • 8 years ago

      Never mind overclocking. Without any overclocking the performance of modern processors is gated by cooling. Vanilla chips in stock motherboards speed up and slow down in response to temperature. Cooling is as important as the grade of chip yet it is rarely a selling point because it is so hard to verify. You can tell what chip you have easily. You can even shake the exact part number out of vendors before you hand over your money. Good luck nailing down the cooling quality.

      I do not buy k series or over clock. I care about reliability and features like the new memory transactions and virtualization more than fine points of clock speed. I do however care about cooling. It is far more important than the inattention it receives. Ivy Bridge running hot is a big detriment.

      • Bensam123
      • 8 years ago

      It’s not about everyone OCing it, it’s about them somehow rationalizing saving a fraction of a sent on shitty thermal paste. It’s like the guy that skimps on a decent PSU for their new build.

      They’d probably save the same amount of money by making the box cheaper. So obviously there is something else going on here.

        • Krogoth
        • 8 years ago

        You would be surprise how bean counters can influence PHBs to cut-corners without the masses noticing it even if just the company a few cents per unit.

          • Airmantharp
          • 8 years ago

          Every extra cent they save is another cent for the bottom line!

          (and that’s just good business)

          • Bensam123
          • 8 years ago

          Yeah and I could understand when it comes to a box or something, but not their own processors…

    • ronch
    • 8 years ago

    I’d love to know just what Intel has to say about using thermal compound with mainstream Ivy Bridge and Haswell models. Is it cheaper? I guess it’s all part of raking in more profits.

      • Airmantharp
      • 8 years ago

      The less-sinister perspective that I don’t really want to acknowledge is that Intel tested each of them with both and found it to not make enough of a difference to justify the cost.

      Intel probably found that it might make some difference for overclockers willing to push the voltage and attach ‘proper’ cooling, but the cost for a divergence in production lines was probably at the core of nixing that.

    • not@home
    • 8 years ago

    [quote<]a change that was in part blamed for the chip's high operating frequencies when overclocked.[/quote<] Shouldn't it be temperatures and not frequencies?

      • willmore
      • 8 years ago

      Sorry, dude, we’re too late with this correction, see excession’s post above.

    • Flatland_Spider
    • 8 years ago

    I’m disappointed that we’re talking about a six core CPU instead of one with cores in the double digits.

      • chuckula
      • 8 years ago

      In binary it has 110 cores, so that’s triple digits!

      • ronch
      • 8 years ago

      Do your workloads benefit from a CPU with more cores than one with higher per-core performance?

        • chuckula
        • 8 years ago

        I know that Flatland has some Linux workloads… I could see situations where a crapton of cores would be a good thing (even if per-core performance isn’t record setting).

      • 0g1
      • 8 years ago

      Yeah, its disappointing the lack of improvement over Sandy-E.

      • NeelyCam
      • 8 years ago

      Get a Xeon Phi, and you’ll be swimming in cores

      • Klimax
      • 8 years ago

      You want some? See Xeons… (Also there are not many tasks which can be so much parallelized…)

      • Oscarcharliezulu
      • 8 years ago

      You could buy a Xeon

    • chuckula
    • 8 years ago

    Hrm… well the die shape is all wrong for a consumer grade Ivy/Sandy/Haswell part. If it’s fake it could have just been a delidded Sandy-E that is already known to have a soldered IHS. Assuming that this is real, it’s good to see Intel is still soldering the IHS for the high-end chips, now they just need to bring it back for desktop parts.

      • Pholostan
      • 8 years ago

      But is the die shape wrong for Ivy-E? I don’t know.

        • chuckula
        • 8 years ago

        One thing that I don’t have time to do: Using the known size of the LGA2011 PCB and some pixel counting, you could get a pretty close estimate of the die size from those photos. You could compare that die size to the SB-E die to figure out if it is the new die or the old die (unless Intel somehow decided to make the IB-E die almost the exact same size as the SB-E die).

    • ronch
    • 8 years ago

    [quote<] a change that was in part blamed for the chip's high operating frequencies when overclocked[/quote<] Shouldn't high frequencies be a good thing when overclocking? Or is that supposed to be ' temperatures'?

    • Bensam123
    • 8 years ago

    Product segmentation at it’s finest…

    When you’re producing super intricate CPUs and you decide to use crappy thermal paste, something is amiss.

      • jihadjoe
      • 8 years ago

      Didnt someone test the paste and find that the stuff intel used is among the best in the business? I recall the intel paste beat some very high end paste. The crappy temps are caused by the adhesive making a gap between the chip and the heat spreader.

        • chuckula
        • 8 years ago

        You’re right. As someone who has delidded a 4770K, I know first hand that the thermal paste isn’t the major issue. The real issue is that there is some variation in the small gap between the IHS and the die due to the adhesive that holds the IHS to the PCB. The actual TIM that Intel uses is pretty high quality, but it doesn’t matter that the TIM is high quality if small air bubbles can form between the die & IHS.

        • Bensam123
        • 8 years ago

        Yet it didn’t have this issue with the SB models with different thermal paste?

        Some substances are better at conducting heat over longer distances too, say metal over paste.

    • JustAnEngineer
    • 8 years ago

    [quote<] They're even toastier when pushed, and overclockers not impressed.[/quote<] Krogoth needs a verb in this sentence.

      • derFunkenstein
      • 8 years ago

      I LOLIRL’d

      • Firestarter
      • 8 years ago

      overclockers collectively krogothed

    • Farting Bob
    • 8 years ago

    Hey guys we know we changed the solder to save ourselves a microcent for a $300 CPU we sell and we know that heavily effected your temperatures but it’s ok cause we’re bringing it back! All you have to do is buy a new expensive CPU. And a new expensive motherboard. See, we totally listened to you!

      • Scrotos
      • 8 years ago

      It’s still better than Microsoft and the start menu! At least you can still buy the solution you want from the oem and not a 3rd party!

      • Airmantharp
      • 8 years ago

      It’s asinine, but it’s working on me.

      A properly overclocked Ivy-E hex-core, say 5.2GHz+ stable on something like an H100i (or one of those alt-brand dual-140mm versions) is pretty appealing and would provide a pretty decent jump over my 2500k running at 4.5GHz. Two extra cores, six extra virtual cores, some Ivy IPC gains, guaranteed PCIe 3.0 x16 to both slots, sounds good to me!

      Now I just need to save up for the ~$600 CPU, $250 board, and $200 RAM. Will have to put off buying my next [url=http://www.bhphotovideo.com/c/product/892851-REG/Tamron_20_200mm_F_2_8_DI_VC.html<]lens.[/url<]

        • mganai
        • 8 years ago

        Or, just put it off for another year for Haswell-E. 😉 6-8 cores, IPC improvements… only issue is that the only 8 core might be the EE model.

          • DeadOfKnight
          • 8 years ago

          Well then, finally a reason to buy an EE, for those who can afford it. That would suck though.

      • GatoRat
      • 8 years ago

      But, is the total cost a “microcent”? Is the process itself more expensive? Does it cause more chip failures? Does it require high costs every time you retool for a new production line?

        • Airmantharp
        • 8 years ago

        It’s probably not the cost but the return on investment. Intel is bigger on delivering to their shareholders than their customers, as good corporate citizens should be.

          • GatoRat
          • 8 years ago

          If the cost is indistinguishable from background noise, than the ROI argument makes no sense. Intel clearly aren’t idiots, so there must be a genuine rational behind their decision besides “let’s be penny pinchers and screw over customers.”

          While I don’t know the answers to my questions, they aren’t entirely rhetorical. I’m willing to wager that the overall costs of using solder aren’t trivial. But, that disrupts the “Intel bad” narrative.

      • ClickClick5
      • 8 years ago

      But in 10 years, the savings will be over $20!

        • crabjokeman
        • 8 years ago

        And then I can put a “Cut Production Costs” bullet on my resume!

    • excession
    • 8 years ago

    “change that was in part blamed for the chip’s high operating frequencies when overclocked.”

    Did you mean “high operating temperatures”?

    🙂

      • danny e.
      • 8 years ago

      yes

      • carpo
      • 8 years ago

      Damage – please, fix the wrong text pointed by [i<]excession[/i<]! Thanks!

      • Dissonance
      • 8 years ago

      Fixed.

    • Celess
    • 8 years ago

    Ouch

Pin It on Pinterest

Share This