No, you can’t enable Threadripper’s extra two dice

Gather 'round, gerbils, I've got a tale to tell. See, before AMD's Ryzen Threadripper released, notorious hardware destroyer der8auer ripped apart a chip and found what appeared to be four silicon dice under the lid. This interesting bit of info arrived shortly before AMD said that Threadripper made use of only two eight-core "Zeppelin" dice, and that the two extra bits of silicon were non-functional "dummies."

Image: der8auer, via Youtube

Despite AMD's statement regarding its "dummy dice," some folks took der8auer's early delidding efforts as evidence that Threadrippers were merely AMD Epyc CPUs by another name. That may have seemed like a reasonable enough conclusion to make. After all, Intel's Core X processors are essentially market-segmented Xeons. der8auer himself could have been among those folks, because he went the extra mile and stripped the substrate from all four dice on another Threadripper CPU that he bought at retail to find that they were in fact fully-etched Zeppelins.

Despite the appearance of functional silicon on all four of Threadripper's dice and their superficial similarity to Epyc packages, AMD maintains that Threadripper CPUs are distinct products from Epyc chips. On Sunday, AMD's James Prior tweeted the following info to that effect:

As Prior says, the two non-functional dice in a Threadripper processor are "basically rocks" with "no path to operation." That's to say that even if they were working processors, there are no connections to them in the Threadripper package. That also means the extra dice in the Threadrippers shipping today can never be enabled through some kind of BIOS hack, dashing the hopes of some optimists who thought they could turn them on somehow.

While der8auer's findings might seem to contradict AMD's earlier statements regarding the nature of these dice, the revelation ultimately makes good sense. AMD's yields on the Zeppelin die are purportedly better than 80%, but that still leaves a whole pile of nonfunctional dies to use as spacers in Threadripper CPUs. It's nice to have confirmation on the exact nature of the extra dice, so hats off to der8auer for his hard work.

Comments closed
    • BIF
    • 2 years ago

    So when did we start calling a “die”, “dice”, and why?

    Lots of words have more than one meaning. For example, a “tap” could be the tool opposite of a “die” in wood or metal shop, or it could be the thing that lets us get beer out of a keg.

    But a CPU die (or the plural, “dies”) was usually not mistaken for a metalworking tool or for the deceased.

    A “joint” has numerous meanings, but if you smoke one then you probably should avoid using the “jointer” in my woodshop (you know, to make the kinds of joints that aren’t smoked) . BTW, shop rules: No leaving of body parts in/on the equipment, or laying around on the floor, and everybody cleans up his or her own blood before leaving!

    I’ve also never confused a wood plane with the bow plane of a boat, a winged aluminum tube, the vertical one at the front edge of the goal line on a football field, or the 3D construct by the same name.

    And besides, dice are already used in numerous applications, although I do think it’s a good thing we Americans don’t hang fuzzy ones on our rear view mirrors anymore. Well, not usually. 😉

      • Mr Bill
      • 2 years ago

      I would have said its a deliberate confusion of terms to make puns. Dicing the wafer into individual dies may lead some wags to say the chip dies are dice when they are merely diced. But the author seems to be saying dice is the plural of dies at least when they are packaged as functional subunits.

    • AMDisDEC
    • 2 years ago

    LOL, trying to get something for nothing, or a free handout, like a GOP conservative.
    Those die are flawed and therefore rendered useless.

      • BIF
      • 2 years ago

      A “free handout?” what are you talking about? A free handout is when you go and take something from somebody, usually at the point of a government gun.

      Nice try, but this is not that!

      Finding a way to tune or enhance your factory car (say opening up the throttle body, adding nitro, putting on racing tires or fatter brakes/rotors, tuning the exhaust, etc) is “hotrodding”, which is in fact, The American Way, born of good old fashioned American ingenuity!

      Have a nice day and thanks for playing the Tech Report home game!

        • AMDisDEC
        • 2 years ago

        A Free-handout is receiving something for nothing. Something that you aren’t entitled to but expect nevertheless.
        Like Donald Trump, the world’s biggest dunce (Dotard?) receiving the free-handout of being elected President, solely on the lack of color of his skin, and Donald Trump Jr. receiving a free-handout for the same reason.

        You purchased a CPU with so many core, yet you expect a free-handout to receive more than you purchased, as if AMD wouldn’t be completely aware there would be people like you who tried to get more than you paid for.
        The game is only in your mind.

        • K-L-Waster
        • 2 years ago

        Except in this case it’s more like buying a 4 cylinder and expecting that the factory secretly slipped a V8 in there instead and just disabled half of them… and all you have to do is that one weird trick to access all the ponies.

        The computing equivalent of hot rodding would be overclocking, case mods, custom liquid cooling loops, etc., not expecting that your ThreadRipper is actually a full Epyc, shhhh, don’t tell anyone….

          • AMDisDEC
          • 2 years ago

          Cadillac once tried that 4-6-8 cylinder gimmick.

    • Kraaketaer
    • 2 years ago

    Finally, a tech news site that knows what the correct plural form of ‘die’ is.

      • ronch
      • 2 years ago

      I still can’t get over associating flat pieces of silicon with cubes though, so I don’t know if I’ll stop using ‘dies’ just yet.

      Well, maybe when Intel dies but until that day…

        • BIF
        • 2 years ago

        I won’t stop.

    • NeelyCam
    • 2 years ago

    Ok. So, a 32-core 64-thread version is coming out by the end of the year.

    You heard it here first, folks.

      • chuckula
      • 2 years ago

      October launch CONFIRMED!

        • ronch
        • 2 years ago

        The Confirmator has spoken.

    • UberGerbil
    • 2 years ago

    [url<]https://youtu.be/6U8JlcB_BzA[/url<]

    • Waco
    • 2 years ago

    I’d honestly be pretty surprised if AMD decided to run multiple substrates for this. That’s a lot of extra cost for essentially no purpose…

    • deruberhanyok
    • 2 years ago

    enthusiasts to AMD: “Can we enable these extra two dies?”

    AMD: “No dice.”

      • BIF
      • 2 years ago

      Okay, that was funny.

    • tipoo
    • 2 years ago

    Is dice for real the proper plural of die, not dies?

      • maxxcool
      • 2 years ago

      Niether, when I am murdering my gaming group I let them know i’m rolling 15 ”die” 20 for damage with a 1d3 multiplier ..

      • ronch
      • 2 years ago

      I think ‘dice’ is changing the word too much. Just add an ‘s’. It’s that simple.

    • Leader952
    • 2 years ago

    These are EPYC processors that failed during testing pure and simple no matter how AMD tries to spin it.

    [url<]https://seekingalpha.com/article/4107952-amd-threadripper-mysteries?auth_param=1dacbj:1cs2c5n:31aac2c3c4a95466bd97c478eef2126f&uprof=44&dr=1[/url<] [quote<]AMD continued to call the non-functional Zeppelins “dummy” dies after the retail Threadripper delidding, according to Tom's: [quote<] We followed up with AMD and inquired if the die were functional. AMD responded that two of the die are in fact non-functional "dummy" die that aren't connected electrically to the substrate.[/quote<] This seems to be dissembling on AMD's part. Reviewers have typically assumed (as did Anandtech) that the “dummies” were mere silicon blanks, not fabricated Zeppelin dies. I find it rather curious that AMD has gone to such lengths to obscure the fact that the dies are fabricated parts, working or not. Tom's concludes: [quote<]Mounting dies to a substrate is like any other manufacturing process; there can be defects. It's possible that the Threadripper processors are simply EPYC processors with die that weren't successfully mated to the substrate. Conversely, it's also possible that AMD is using defective die for the fillers, but in either case, it's doubtful that the company is wasting functioning silicon.[/quote<] In fact I find the former explanation, that Threadripper parts are simply defective EPYC processors, far more likely. The reason is that even using defective Zeppelins for spacers is very inefficient from a manufacturing standpoint. The dies have to be mounted in some fashion as the functional Zeppelins are, and then they are soldered to the heat spreader as shown in der8auer's video. That's a lot of work, especially the soldering to the spreader, for dies that are non-functional and won't generate any heat. Probably a simpler spacer scheme could have been devised, such as machining the spacer into the underside of the heat spreader. [/quote<]

      • ludi
      • 2 years ago

      Regardless of whether this speculation is true, all four dies will need to be interfaced to the IHS the same way to ensure there is no flexing under the varying shapes and sizes of the cooling solution’s interface. If the active two are soldered, then all four will have to be.

      • UberGerbil
      • 2 years ago

      [quote<]Probably a simpler spacer scheme could have been devised, such as machining the spacer into the underside of the heat spreader.[/quote<]What? That sounds more complicated and more expensive. They already have a production line designed to assemble a product from four pieces of silicon on a substrate. It's got to be cheaper and easier to just use a couple of non-functional pieces of silicon in some cases and keep everything else the same.

      • K-L-Waster
      • 2 years ago

      Not sure how much weight to give to an opinion piece from an investing site on this kind of question.

      • ronch
      • 2 years ago

      OK even if these are Epyc then so what? It’s kinda like binning but on a package level then. As long as I get my cores at a certain clock then what’s the issue?

        • K-L-Waster
        • 2 years ago

        The “issue” is people still think there will be an after market way to unlock the extra dies despite AMD saying that nope you can’t….

          • ronch
          • 2 years ago

          I think he’s saying using defective dies deliberately is very inefficient, and he believes the dead dies just weren’t placed on the package successfully. I find that a bit dubious. How hard is mounting a die on a package? Has this been an issue at all before? And using a different material as a rock would’ve required some retooling. Also, doesn’t AMD test dies before mounting them on a package? Mounting every chip from the wafer before testing them to know which ones work and which ones don’t doesn’t seem very efficient to me.

          There’s also an article I posted here about Threadripper being a skunk works project. Based on that article I would think that Threadrippers are essentially Epyc chips, using the same production line, probably, but also makes use of dead dies. AMD probably doesn’t want to retool the production line to include different materials to use as rocks, and the equipment needed to mount dead dies is there anyway so why not just use those dead dies? No retooling, less waste.

      • TheMonkeyKing
      • 2 years ago

      oh, yeah? Well, I’ve got a lead pencil that’ll get that fixed.
      #Athlon4Ever

      • Waco
      • 2 years ago

      You’re wrong, no matter how you try to spin it. 🙂

      Do you really think AMD is building packages with four dies (of which two on opposing corners MUST work for TR) and taking a gamble on that assembled package just working?

      The dies are tested *prior* to package assembly. There’s no way the assembly rate is so low that they can make an entire product line out of the ones that *happen* to have the correct two dies dead on assembly.

      This theory makes no sense.

    • jihadjoe
    • 2 years ago

    This really changes the estimate on AMD’s yields. When Ryzen came out we all thought that yields were awesome and AMD was getting fully-functional 8-core dies left, right and center. Now that Threadripper’s undies have been exposed it appears that might not be the case at all.

      • Wirko
      • 2 years ago

      Yes, the new estimate is < 100%, which is quite disappointing.

        • chuckula
        • 2 years ago

        AMD yields, my dear friends, are 93% perspiration, 6% electricity, 4% evaporation, and 2% butterscotch ripple.

        That’s 105 percent!

        • ronch
        • 2 years ago

        Oh wow. What a terrible waste of silicon to throw away even just one die… T_T

      • willyolioleo
      • 2 years ago

      do you really believe Threadripper is a high-volume part?

      • just brew it!
      • 2 years ago

      What are you smoking? Even if they’re getting close to 100% yields now, they’ve probably got enough defective Ryzen dies already stockpiled to provide them with Threadripper spacers for months, if not years. I expect volume of Threadripper will be an order of magnitude less than Ryzen.

        • DPete27
        • 2 years ago

        This is the only logical explanation I can think of as to why they’d use actual dies as spacers, especially if their yields are as high as they claim.

          • Wirko
          • 2 years ago

          Maybe these un-dies were rejected at an earlier stage of manufacture (as un-wafers?), and don’t even count as un-yield.

        • ronch
        • 2 years ago

        I actually find it very comforting that they found a real use for those throw-away dies that also actually makes it easier for them to manufacture these things without having the need to retool because they wanted to use a different material as a rock. If those were real working dies then it gives me much grief but if they’re destined for the bin (get it?) anyway then yeah, might as well use them as rocks in there.

    • Pancake
    • 2 years ago

    “No, you can’t enable Threadripper’s extra two dice”

    DON’T TELL ME WHAT I CAN’T DO!!!

      • ronch
      • 2 years ago

      OK I won’t, but you still can’t.

    • Tumbleweed
    • 2 years ago

    Threadripper dice: I always roll eights.

      • jihadjoe
      • 2 years ago

      4D8

    • ronch
    • 2 years ago

    Of those 4 dies, are the locations of the 2 working dies always the same? Or is it er… a roll of the dice?

      • SiSiX
      • 2 years ago

      The active dice are on the diagonal, same as the dummies. Done for heat management reasons. (And I would imagine also for hookup/communication purposes as well. )

    • ronch
    • 2 years ago

    I’m hoping those dead dies are dies that were really DOA in the Fab which AMD might as well use as rocks, otherwise it’s a waste of silicon of Epyc proportions.

      • BIF
      • 2 years ago

      Okay, I liked that. No negs for you, but I have given all I can…

    • Wirko
    • 2 years ago

    Every piece of silicon is basically a rock. I mean, there’s still hope.

      • tipoo
      • 2 years ago

      Whenever my code feels hacky I remind myself we just tricked rocks into thinking

        • just brew it!
        • 2 years ago

        By feeding them 0s and 1s, no less. Only two unique digits! How hard can coding possibly be? 😉

          • UberGerbil
          • 2 years ago

          Especially with object-oriented programming. Just create a no-bugs class and inherit from it for every object.

            • just brew it!
            • 2 years ago

            Sounds like that must be a variant of the DWIM (Do What I Mean) class?

            • Firestarter
            • 2 years ago

            just have it implement the MindReader interface, BAM you’re done

            • [+Duracell-]
            • 2 years ago

            I just puked…thanks.

      • p3ngwin
      • 2 years ago

      Silicon is a mineral, not a rock.

        • Froz
        • 2 years ago

        As not a native speaker, I have to ask – what exactly is wrong here? Aren’t rocks made of minerals?

          • tipoo
          • 2 years ago

          Yep. Further, silicon is largely derived from sand, aka finely divided rocks.

        • just brew it!
        • 2 years ago

        If we’re going to be pedantic, actually it is classified as a metalloid.

        • BIF
        • 2 years ago

        I just put a new back cover on my iPhone 6+. Made of silicone, it doesn’t feel at all like a rock. But it does seem to be about as dumb as one.

          • e1jones
          • 2 years ago

          I’m a mechanical engineer. Had to deal with another mechanical engineer that used ‘silicon’ and ‘silicone’ interchangeably. Which was really annoying when both were topics of conversation depending on the project we were discussing.

          Also, he was thought his $hit didn’t stink.

    • seeker010
    • 2 years ago

    core dice…. did I miss a memo somewhere?

      • thedosbox
      • 2 years ago

      AMD’s new marketing idea – roll the dice, that’s how many cores you get.

      • ludi
      • 2 years ago

      He’s play on a typical English plural construction. In gaming, a multi-faceted object with numbers or characters on each face is a “die,” while two or more are “dice.”

      For industrial processes, a device that creates or retains a physical impression of something else is also called a “die,” but the plural is “dies.” Unless it involves colors, in which case we have dyes.

      Not to be confused with the cessation of life processes, of course.

      Did you hear about the woman named Diana who accidentally killed her CPU collection when she dropped it in a vat of coloring agent that she was using to modify her gaming collection? Di’s dies die while Di dyes dice.

        • just brew it!
        • 2 years ago

        IIRC either form is considered correct when dealing with silicon dies/dice.

          • seeker010
          • 2 years ago

          I had a professor who told me, “If someone calls dies dies, you know they went to a real university (aka an American one) and got a real degree. If they call it dice they are probably skinny jean wearing, chai tea drinking, European wannabes with neckbeards who spent a semester in Europe thinking they saw the entire world and got tempted to drop out (although maybe he was thinking of the term hipsters). And if they call it die they are just ignorant SOBs and you should have nothing to do with them.”

          This was years ago and yet I’ve still kept the prejudice. Goes to show if you’re not in the field, someone else’s opinion can really color how you perceive things depending on when you heard it. Also you’re welcome for the new mental picture of the author.

        • bthylafh
        • 2 years ago

        Purportedly.

        • tipoo
        • 2 years ago

        >Did you hear about the woman named Diana who accidentally killed her CPU collection when she dropped it in a vat of coloring agent that she was using to modify her gaming collection? Di’s dies die while Di dyes dice.

        Jesus, lol.
        Community award.

    • CScottG
    • 2 years ago

    der8auer blows on processor and yells:

    “Baby needs a new pair of shoes”, and then

    “..awe, crap: Snake Eyes!”

      • BIF
      • 2 years ago

      No, it would be more like, “aww crap; it was tails!”

    • tipoo
    • 2 years ago

    Does “no path to operation” mean they were never tested and deemed failed? So why are the dies etched at all? Or can they test without that, which would make the etching make sense?

      • derFunkenstein
      • 2 years ago

      I took it to mean they were tested, marked as failed, and then stashed for their day of glory where they were just glued into place so that the IHS had something with which to make contact on those positions.

        • crystall
        • 2 years ago

        They’re probably soldered to the package rather than glued, this would give them the same structural behavior of the other dies. It would also allow AMD to place the good and dead dies on the package in one go using the same process they use for good dies.

          • derFunkenstein
          • 2 years ago

          Yeah that probably makes sense and probably explains why some folks thought they might be accessible and just dormant.

      • ColeLT1
      • 2 years ago

      They could be perfectly good chips (or failed some test), but no connections in the package to any pins or wiring, but the actual die is the real deal.

      • chuckula
      • 2 years ago

      I’m assuming they [b<]were[/b<] tested and deemed failed then slapped onto the substrate to act as a physical support member to hold the huge heatspreader on the package. In a multi-chip monster like threadripper they don't just randomly throw dice onto the package and hope that they work. Instead, each die is tested first, put on the package, and then the full package is presumably tested again to make sure the packaging process itself didn't introduce a failure. Put another way, even with the supposedly "great" 80% yield number* if AMD didn't test each die individually before it was put on a 4-die Epyc package then only 41% of the Epyc packages that come off the assembly line would actually be functional (0.8^4 = 41%). * At Intel claiming that you have a yield of 80% on a mass-produced part outside of something really high-margin and exotic would probably get you fired. Claiming a 90% yield might still get you fired. Claiming well above 90% is more like bragging that the lights are working in the men's room: They better work.

        • Goty
        • 2 years ago

        A [url=http://www.barrons.com/articles/amds-ramps-epyc-wafers-yields-to-steal-intel-server-biz-says-arthur-wood-1505767497<]Barron's article[/url<] (warning: paywall) suggests yields are likely significantly greater than 80%. Not sure how much faith to put in that claim, though, since I'm not familiar with the firm making the statement.

      • Helmore
      • 2 years ago

      They were probably using actually etched but nonfunctional dies as those would have been thrown away otherwise anyway. Once yields improve I expect that to change to them just using appropriate spacers.

      • just brew it!
      • 2 years ago

      I take it to mean “they’re physically present but not electrically connected”.

      They will probably use only etched (but defective) dies if they have enough defective dies to do that, and start inserting unetched chips of the same size if they run out of defective ones.

      Even if they’re getting 80% yields they should have plenty of bad ones to use for Threadripper, as I’ll bet volume of Threadripper will be fairly small compared to Ryzen.

      • BIF
      • 2 years ago

      And the add-on question: Do the failed dies always end up in the same position, and what if they don’t…doesn’t that still tend to raise the cost of manufacture?

    • chuckula
    • 2 years ago

    Arrrgh! Ye scurvy landlubbers’ll never get me non-functional Ripper Dice!

    X on the Heatspreader marks the spot see!

      • derFunkenstein
      • 2 years ago

      But if I use the world’s tiniest soldering iron, can I make functional Ryzen CPUs out of them?

        • southrncomfortjm
        • 2 years ago

        Sure. Yes. Maybe. Go for it.

Pin It on Pinterest

Share This