back to article Nvidia slowed RTX 4090 GPU by 11 percent, to make it 100 percent legal for export to China

Nvidia's GeForce RTX 4090 GPU is back on sale in China – in a less capable configuration version designed to comply with US restrictions on exports into the Middle Kingdom. Dubbed the RTX 4090D, the device appeared on Nvidia's Chinese-market website Thursday and boasts performance roughly 10.94 percent lower than the model …

  1. An_Old_Dog Silver badge

    Slower Version

    This makes me think of the -- urban tech legend? -- of some of IBM's various, compatible mainframe computers being "the same" inside, and that when a customer paid for an upgrade to a higher model, a tech would come out, install a jumper, and laze around pretending to work for an hour or two. The advantage to IBM (or any other manufacturer) of such a scheme is having to have only one production line for all of its compatible models across a CPU speed range. The disadvantage to IBM (or other manufacturers) is the customer or some hacker might learn of this, and install a "free" speed upgrade for themselves. These days, if such a scheme exists, it probably involves a BIOS change.

    I can't help but wonder if nVIDIA has some magic sauce involving test pins and/or JTAG to enable all the cores, and raise the clock speed, of its chips, thus changing the crippled "export" model into the "regular" model.

    1. Anonymous Coward
      Anonymous Coward

      It will be in the firmware

      NVIDIA has form for crippling cards using firmware changes, that’s how they historically separated Quadros from their GeForce lineup. Enforcement this time round will totally be done using digital signature checks plus natural binning (e.g. if a card has some defective cores, ensuring they are not used but shipping the card anyway, as the spec demands fewer working cores anyway).

      In the event someone does defeat the restrictions, they’re left with a card which is statistically less likely to work properly anyway, allowing for legitimate plausible deniability, and if any shenanigans happen, the US government would not want to lose face, nor risk destabilising stocks/shares for which and wealthy investors. So nobody will admit anything and the CCP would probably play along even when their people do find a way round it by not shouting from the rooftops, just continuing to buy more.

      1. cyberdemon Silver badge
        Holmes

        RTX4090D

        Does the D stand for Dud or Duff?

        What to do with this bin of chips where 10% of the cores failed validation?

        Perfect idea!

    2. Dickie Mosfet

      Re: Slower Version

      I'm sure I heard somewhere that the earliest IBM AT PC's, which shipped with a 6MHz 80286, could be given a very cheap speed bump by replacing the 6MHz clock crystal with an 8MHz equivalent.

      Once IBM heard about this, either the design or components were changed to prevent this from working. IBM weren't about to let anyone have a quicker computer without paying for it!

      1. david 12 Silver badge

        Re: Slower Version

        The first IBM AT had a 6MHz bus, designed with pre-release Intel 80286 parts. When it became clear that their existing design could be fitted with an 8MHz processor and overclocked to 8MHz, their response was to release the 8MHz version. The 6MHz version was never stable just boosting the clock speed: Intel shipped 6MHz processors defined by "can't reliably do 8Mhz".

        1. Dickie Mosfet

          Re: Slower Version

          Thank you for the clarification, David!

    3. aerogems Silver badge

      Re: Slower Version

      I've also heard tale of how companies will "repurpose" chips that may not quite pass muster. So, everything starts out life as the top of the line 4090, but during QA some of them might have some minor defects. Like not enough of the tensor cores are fully functional to meet the requirements of the 4090, but otherwise everything works perfectly, so they rebadge it as a 4080. Then some may not meet the specs for 4080 and become a 4070 and so on until either all the chips have been rebadged or you run out of models and scrap the rest. I think this is how AMD came up with their old tri-core CPUs. They were quad-core designs, but one of the cores didn't pass testing, so they disabled it and sold it as a tri-core CPU.

      1. Anonymous Coward
        Anonymous Coward

        Re: Slower Version

        Yep, this is pretty well known. You'll often see it referred to as "binning". Defects are so common that it doesn't make sense to discard hardware that's 99.9% good. Though, I think manufacturers still tend to have a couple product lines that they try to manufacture, not just one.

        1. ldo Silver badge

          Re: Slower Version

          Like, how about the Intel 486SX? The 486 chip had onboard floating-point. But if this was defective in a manufactured chip but the rest of it was OK, it was sold as a “486SX”. The motherboards with these had an empty spot for another chip you could buy, called the “487SX”. This was actually a full-function 486 that disabled the 486SX and took over its functions.

      2. Phil O'Sophical Silver badge

        Re: Slower Version

        Didn't Clive Sinclair start off in business by buying bulk lots of transistors that failed to meet the specs like hFE, retesting them, and selling them to the hobbyist market as lower-performance devices.

        1. Alan Brown Silver badge

          Re: Slower Version

          Clive wasn't the only one who did that

          "Back in the day", transistor manufacturing was very hit-and-miss in terms of the chemistry and two "identical" batches would have quite different characteristics (and a lot of variation within each batch too)

          In most cases the batch was measured than then given a type number which matched the results, or a new type number created

          1. david 12 Silver badge

            Re: Slower Version

            "Back in the day", my old man had a story about an early MIL-SPEC transistor, (when "clean rooms" were new thing): they needed to settle/ fix/ define/ sanction/ document the process, so they supervised catalogued and recorded everything. And the process stopped working, and they couldn't fulfill their high value MIL contract.

            So they got everyone together and told them "no consequences, we'd really like to know what you were doing wrong when it was working, so that we can keep on doing that".

            And then paid some of the staff to hide behind the ovens and smoke cigarettes, to get the semi-conductor defect level up where they wanted it.

      3. Anonymous Coward
        Anonymous Coward

        Re: Slower Version

        Those tri-core CPUs were pretty good at the time if I recall. I had one, think it was a Phenom and it did me well for a while.

      4. katrinab Silver badge

        Re: Slower Version

        That is not a tale, it is standard practice. Intel for example doesn’t make about 5000 different CPU models, they make a few, and bin the resulting production based on QA test results.

      5. Annihilator

        Re: Slower Version

        Yep. Most if not all CPUs were manufactured in this way, especially in the earlier days when all that defined a CPU was its base clock speed and multipliers. Essentially all Pentiums were created on the same die, tested to see how it performed and then badged at the correct clock speed. The higher cost was primarily down to the rarity and low yields - a P200 was identical to a P133 off the production line, just one was more stable than the other.

        Similarly with cores as you mentioned, they'd sell 3-core processors. Realistically, no one was ever going to design a 3-core CPU, but a defective core on a quad-core device was fairly common. AMD did this with the Phenom range - Toliman (3-core Phenom) was an Agena chip with a defective core.

    4. Anonymous Coward
      Boffin

      Re: Slower Version

      > This makes me think of the -- urban tech legend? ..

      After asking people to leave the room, a tech would take the panel off the printer and move a belt up a notch, that doubled the print speed.

    5. Yorick Hunt Silver badge

      Re: Slower Version

      Dell's hobbled iDRAC which magically springs to full life when you enter an enterprise licence key in the BIOS? BMW's heated seats? There're many examples.

      1. Anonymous Coward
        Anonymous Coward

        Re: Slower Version

        There have been ways around iDRAC licensing for a long time.

        I remember a time where keys were boy d to the service tag of the machine and the machine allowed you change the service tag in the BIOS. So you could activate many servers with the same key then set the service tag back.

    6. abend0c4 Silver badge

      Re: Slower Version

      When Digital introduced the MicroVAX it became clear that they needed more than one model (for market segmentation) but that the next generation CPU was some way off.

      They paid a bloke to take a certain number of the motherboards off the production line and fill the expansion slots with epoxy resin. These formed the basis of the "entry level" version despite costing a little more to produce. The wonders of marketing.

    7. Giles C Silver badge

      Re: Slower Version

      A few years ago (about 7) I am sure that the IBM system-I that my employer ran used the same system at a performance tier, if you wanted more power then they would install a license allowing you to use extra cpu cores. I think you could do it on a time limited basis I.e. need more power for job x buy a license for two weeks and then go back to the slower level

    8. ridley

      Re: Slower Version

      It's not an urban legend.

      I've seen what the IBM tech did no increase the speed of our printer at great expense, and it didn't involve any new parts and took seconds.

    9. Zibob Silver badge

      Re: Slower Version

      That is however exactly what BMW did and people hated it. So it does for sure happen.

      I can also think of FLIR Esomething having a higher grade sensor regardless of model bought. And it was something silly like a text file telling the device what it was.

      Possibly also a multimeter, or bench top Oscope. All the same. Ship the full beans and limit the user to whatever. Usually gets caught out soon enough.

      1. CountCadaver Silver badge

        Re: Slower Version

        Also many (but not all) cars sold with differing hp outputs on the same engine are different only by engine mapping (VAGs 1.2 TSI 90 and 110 could both be mapped to 130 - yet people tried to shout me down online that this was nonsense....someone waded in who mapped cars for a living and made it very clear how prevalent this practice was)

    10. CountCadaver Silver badge

      Re: Slower Version

      Could also be a physical break in the connections or using cards which failed verification at full speed

  2. HuBo Silver badge
    Mushroom

    Easiest limbo ever

    Glad to hear it was an easy fix! (as a bonus, this may help prevent the power plug from melting!)

  3. ldo Silver badge

    “I am going to control it the very next day”

    So set a threshold, and then get upset when companies sell products designed to just get under the threshold. So then you want to lower the threshold to ban the new products? Then are you going to get upset again when they rejig their products to just get under that? You wanna play Export Control Limbo Rock? How low can you go? Why not just go there to begin with?

    1. Anonymous Coward
      Anonymous Coward

      Re: “I am going to control it the very next day”

      Gun manufacturers have been doing it for decades at this point. What were they expecting to happen?

  4. Groo The Wanderer

    Funny how the intended result of embargoes and sanctions (blocking the "enemy" from obtaining any access to specific goods or services) is so easily stymied by trivial changes like this. If the goal is to reduce the raw compute capability available to the Chinese, the entire 40xx series should be banned from sale to China. Restrict unwanted nationals to last-generation parts, period.

  5. Michael Hoffmann Silver badge

    So China buys 11% more of them?

    I sadly haven't been able to keep up with that side of HPC as much as I would have liked, so would simply buying 11% not easily make up for it? Just a slightly bigger grid?

    It's not like China cares one iota about their carbon footprint and is stomping out new coal and gas power plants like Xmas cookies. Is having 11% in chip performance that much of a throttle? I'm sure there ElReg commenters who could enlighten me?

    1. HuBo Silver badge
      Coat

      Re: So China buys 11% more of them?

      Right, the throttle is not a lumbago-inducing āsana of recursive yoga sutra, even between H100 and H800 ( https://www.techpowerup.com/gpu-specs/ ). With respect to HPC though (FP64) the difference between RTX 4090 and H100 is huuuge, with the RTX at 1.3 TF/s, and the H100 at 31 TF/s (H800 is still 26-30 TF/s though).

  6. Bebu
    Big Brother

    Fusible link...

    I can imagine these devices might magically regain the the missing CUDA cores with the right incantation if not actually a fusible link. ;)

    Companies with a future wisely look to a future where the middle kingdom is potentially the largest market on the planet even if actually living there made the world of '1984' seem benevolent in comparison.

  7. AusMatt

    NVIDIA laughing

    China will buy 10000% more of these and NVIDIA will be laughing all the way to the bank.

    1. Yet Another Anonymous coward Silver badge

      Re: NVIDIA laughing

      At least it's only graphics cards. Imagine if Boeing were allowed to sell jet aircraft to the wily oriental foe

      1. spireite Silver badge

        Boeing....

        Raises the question at what point did Boeing start putting binned components in the 737MAX ?

        They ceratily binned their foibles there.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like