Der8auer delids a Core i9-7920X

Champion overclocker Der8auer is up to his usual shenanigans. Over on Facebook, he's posted up a racy topless image of what he says is a Core i9-7920X. This 12-core CPU shares the same die with the rest of the top tier of Skylake-X CPUs that goes up to 18 cores. These CPUs have hit store shelves just recently, and the man has already popped the lid on one of them to reveal the mirror-smooth silicon beneath.

Image source: Der8auer on Facebook

Der8auer's post was actually intended to inform his followers that the existing Delid Die Mate X tool will suffice just fine to pop the top on the high-core-count LGA 2066 CPUs. The tool is already for sale at shops like Overclockers UK and is similar to my own Rockit 88 delid tool: simply drop your CPU in the correct place, tighten some bolts, and after a satisfying snap you've got a delidded CPU all ready for cleaning.

The perfect finish of the chip below the cap and the lack of solder residue are pretty strong indicators that these CPUs have thermal paste under their lids. Furthermore, der8auer would hardly recommend anyone to delid a soldered CPU and run a strong risk of ruining it. This all means that like the 10-core-and-below models and regular desktop chips on back to Ivy Bridge, higher-end Skylake-X CPUs all seem to have non-soldered heatspreaders. If this info's accurate, it makes us nervous about the out-of-the-box thermal performance of the bigger chips, given that existing Core X CPUs already run awfully warm.

According to some overclockers and my own experience, unsoldered Intel CPUs sometimes end up with an air gap between the die and the heatspreader—though it's unclear exactly why this happens. Delidding the CPU and replacing the goop inside with a bonding compound like Coollaboratory's Liquid Ultra can provide massive improvements in cooling performance. If Der8auer's discovery carries over to most retail CPUs, delidding might be a good idea to get the most out of Intel's top-tier desktop chips. For those looking to get their hands on a Core i9-7920X of their own, you can grab one at Newegg or from Amazon for a modest $1200.

Comments closed
    • DragonDaddyBear
    • 2 years ago

    It just makes me mad to think about how greedy and stupid this is, saving pennies using an inferior TIM on such a high-end part.

      • Anonymous Coward
      • 2 years ago

      Not overclocker friendly, but Intel must have found that this meets some very strict durability requirements, or they would not do it.

        • DragonDaddyBear
        • 2 years ago

        My comment isn’t about the durability. I don’t question Intel’s QA process. My point is it is a high-end, premium part. I think they should use a premium TIM that costs just a little more. Not doing so is pointlessly greedy and hurts their image more than it saves them money.

          • CuttinHobo
          • 2 years ago

          No, you’re getting a layer of Elmer’s glue and you’re gonna like it!

          • psuedonymous
          • 2 years ago

          [quote<] I think they should use a premium TIM that costs just a little more.[/quote<] The TIM is perfectly fine. It's [url=https://forums.anandtech.com/threads/delidded-my-i7-3770k-loaded-temperatures-drop-by-20%C2%B0c-at-4-7ghz.2261855/page-23#post-34053183<]the gap between the top of the die and underside of the IHS that is 100% of the contribution to the 'temperature drop' of de-lidding[/url<] ([url=http://i.imgur.com/DLdU0rE.jpg<]rehost of relevant graph, because Photobucket killed themselves[/url<]). Indeed, 'good' TIM plus the same gap gives [i<]worse[/i<] performance than the stock TIM! As for 'why even leave a gap then?': so if uneven pressure is placed on the IHS it deforms without crushing the die. Soldering to the IHS bonds them rigidly so they do not flex relative to each other, but without the rigid bond you can concentrate stress onto one point on the die and destroy it.

            • DragonDaddyBear
            • 2 years ago

            Very well articulated point. I haven’t thought of it that way. However, I don’t personally know of any one that has cracked a die. As small as these transistors are, I think they can manage to keep the shield close as they have in the past.

            If the performance of de-lidding and using a metallic TIM, similar to solder, increases the performance of cooling solutions on a 180W TDP part I think the risk is acceptable. I also think Intel would believe the same or the would have switched to a gooier TIM a long time ago, like a low-melting-point liquid metal that is popular in these de-lidding projects.

            Honest question: If the de-lidding gets the shield closer then has Intel changed their spec for spacing?

            • psuedonymous
            • 2 years ago

            ‘Liquid metal’ eutectic alloy TIMs are not something I would ever let near a bare silicon die. They have a propensity for infiltrating other metals (the reason why they are banned from any environments that handle structural Aluminium), and with flip-chip production the upper side of the exposed die is the fine-pitched transistor area most vulnerable to changes in doping.

            If you’re going for extreme overclocks and expect a chip to only last a year or less, then that may not be a concern, but if you expect to assemble a chip package and need it to be viable many years later it’s not a risk you can take.

            • chuckula
            • 2 years ago

            I’ve got liquid metal TIM on my 4+ year old 4770K that’s been there since 2013 without any reapplication.

            It works fine for letting the chip OC to 4.7GHz and no destruction to the chip has occurred yet… we’ll see if it makes it to 2020 for my projected replacement.

            • Kougar
            • 2 years ago

            That seems unlikely when Skylake’s thinner package substrate will bend under such pressure and even break before that happens.

    • Antimatter
    • 2 years ago

    Why don’t AMD and Intel sell CPUs without heatspreaders? I assume the exposed chip is easily damaged?

      • Misel
      • 2 years ago

      I remember that a retailer in my area stopped selling socketed Athlon XPs because users damaged them too easily during installation. Back then Intel had the Pentium 4 and they already had heat spreaders.

      • Krogoth
      • 2 years ago

      Yep

      AMD and Intel started using headspreaders when HSF started becoming larger enough to easily damage exposed dies with improper installation or when chassis gets move around.

    • tipoo
    • 2 years ago

    TIM even for a high end part like this. I’d prefer the Threadripper approach of soldering, to avoid the TIM application lottery.

      • DPete27
      • 2 years ago

      [url=https://techreport.com/news/22859/why-is-ivy-bridge-so-hot-and-bothered<]Welcome to 2012[/url<] They only recently started using TIM on their HEDT chips though.

        • RAGEPRO
        • 2 years ago

        Heh, I love that this comment got -20 when he was completely on the money:
        [quote<]That's not really relevant to this article. The TIM instead of solder is the biggest change regarding high CPU temperatures under overclocking.[/quote<]

          • derFunkenstein
          • 2 years ago

          Folks tend to recoil at anything new, so in retrospect we can recognize how ahead of the curve Duck was.

    • Wirko
    • 2 years ago

    What is the second substrate doing here?

      • RAGEPRO
      • 2 years ago

      Making it easier for Intel to use multiple dies on one socket (and multiple sockets for one die).

        • psuedonymous
        • 2 years ago

        For that they could just change the ‘base’ substrate rather than wasting money on an extra substrate (i.e. they still need to create a die-dependant substrate in addition to the die independent substrate, rather than just sub a right-sized die-dependant substrate to start with).

        Either they expect to place the same dies on a physically smaller socket (or more likely, direct solder to a host board), or they want to future-proof for adding EMIB assemblies without having to use such a large host substrate to embed the silicon bridges into. My bet is more on the former, EMIB for a CPU is probably not going to happen until a new HEDT socket generation anyway.

          • RAGEPRO
          • 2 years ago

          [quote<]Either they expect to place the same dies on a physically smaller socket[/quote<]This is the part where I know you didn't read my whole post, heh. LGA 2066 plays host to at least three separate dies, and at least two of those dies are used on other sockets including the smaller LGA 1151 socket. To be clear, we actually agree, but the tone of your post is as if you're correcting me when you just said the same thing I did with a lot more words.

            • MOSFET
            • 2 years ago

            With substrate-gate going on here, it almost looks like an 1151 CPU on socket steroids.

      • Mr Bill
      • 2 years ago

      Interposer? Only its on the outside, er exerposer?

      • chuckula
      • 2 years ago

      What second substrate?

        • tay
        • 2 years ago

        Look closely it looks like a chip mounted on a chip. It’s chips all the way down and intel labs.

          • chuckula
          • 2 years ago

          I have no idea what people are talking about. It’s clearly a chip mounted on a PCB just like every other standard CPU.

          Are you talking about the potting material around the chip? That’s not any kind of “substrate” that’s just an ordinary component that helps to protect the chip against physical shock.

            • Redocbew
            • 2 years ago

            I believe that was a [url=https://en.wikipedia.org/wiki/Turtles_all_the_way_down<]turtles all the way down[/url<] reference.

            • Spunjji
            • 2 years ago

            Take another look – it goes:
            Die > brown potting material > PCB > black potting material > PCB

            There are also two bands of black adhesive, some on the “upper” PCB and some on the “lower”.

            • Kougar
            • 2 years ago

            The PCB color doesn’t even match. The shiny black gloss looks to be epoxy resin where the PCB is fused to the second PCB. The gold triangle and circles in the corner also denotes orientation of both PCB substrates.

            • Mr Bill
            • 2 years ago

            Oh, then this is the back of the chip? All the package connections probably have to be made on the other side of the chip. Maybe like it is [url=https://electronics.stackexchange.com/questions/56649/what-is-a-die-package<]here?[/url<]

            • Mr Bill
            • 2 years ago

            Old CPU but like [url=http://www.instructables.com/id/How-to-expose-the-die-contents-of-an-old-CPU/<]this.[/url<] And especially like [url=https://upload.wikimedia.org/wikipedia/commons/0/02/80486dx2-large.jpg<]this close up shot[/url<] from that same page.

            • Mr Bill
            • 2 years ago

            I think you have it right. Its just that the what is visible is the back of the chip not the side that has the connections. It would be pretty cool to see how that assembly is laid out is on a modern CPU.

        • Wirko
        • 2 years ago

        It’s easy to focus on the giant brick of silicon and unsee what’s around it. There’s a green substrate with a “001” marking and some goo on it and a larger green sub-substrate with a “R4” marking and some more goo. The whole thing looks to me like a trap set up for the delidders.

          • Anonymous Coward
          • 2 years ago

          They think they can see the die, but its a [i<]decoy[/i<]. The real die is still hidden from view.

    • southrncomfortjm
    • 2 years ago

    Seems to me that Intel should have delidded SKUs up for sale and save people the effort if they aren’t going to apply proper thermal compound to their high end chips.

      • Wirko
      • 2 years ago

      Indeed.

      Hmmm, [url=https://en.wikipedia.org/wiki/BMW_M<]M9[/url<] could just be the perfect brand name for those.

    • DPete27
    • 2 years ago

    remind me again. Do people usually delid, then just squirt TIM on and cover with their water block? Or is there some reason to re-apply the heatspreader?

      • curtisb
      • 2 years ago

      Because all of the heatsinks are made expecting the heatspreader to be there. If it’s not, you’ll end up with a gap between the die and the heatsink surface. There’s also a chance that you’ll crush the die on your $1200 CPU.

      Edit: Typo

        • ColeLT1
        • 2 years ago

        Not only that, but the heat spreader is what the motherboard latching mechanism attaches to, it grabs the spreader, not the chip.

        I ran ivy bridge and haswell die-on-waterblock (no heatspreader) with EK’s “naked” kit, and it ran extra cold. That ended because the PCBs are too thin on skylake on, and pressing the center of the chip (vs grabbing the outside of the spreader) causes it to flex in the middle, making contact with the pins in the middle but not on the outside.

    • brucethemoose
    • 2 years ago

    The larger die should actually help the 7920X dissipate heat, right? More surface area/less heat density.

    Of course, that doesn’t help the VRMs, and all bets are off with the higher core count chips.

      • smilingcrow
      • 2 years ago

      It’s not just a larger die but also more cores/cache etc so overall there may not be much difference.

        • NoOne ButMe
        • 2 years ago

        these chips seem to be all running into the same TPD wall, so at stock, the larger die will help.
        And when overclocking it should, in theory….

      • Goty
      • 2 years ago

      Lower heat density implies the same heat production spread over a larger surface, which I don’t think is a very good assumption.

    • chuckula
    • 2 years ago

    Incidentally, that die is estimated to be [url=https://techreport.com/forums/viewtopic.php?f=2&t=120049&view=unread#p1361588<]491mm^2 or about 50% larger than the "small" Skylake X dies[/url<].

Pin It on Pinterest

Share This