PCIe 4.0 won’t actually deliver 300 watts from the slot

A few days ago, we reported on a story from Tom's Hardware about the power on offer from PCI Express 4.0 expansion slots. Tom's had been in touch with PCI-SIG, the governing body of the standard, and the site reported that PCI Express 4.0 would not only double the throughput of PCIe 3.0, but that it would also allow expansion cards to draw "a minimum of 300W" from the motherboard.

As it happens, the data on PCIe 4.0's upgraded bandwidth was accurate, but apparently someone at PCI-SIG gave Tom's incorrect information regarding the power provisions in the new standard. Currently, PCI Express 3.0 cards may draw 75W from the expansion slot, and an additional 225W from auxiliary connectors, for a total of 300W. PCIe 4.0 devices may be allowed to draw more than 300W in total power. However, the power provided by the slot itself will remain at 75 watts. Folks hoping for a slot-powered GTX 1180 or Radeon Fury RX will probably have to keep dreaming.

Comments closed
    • mcnabney
    • 3 years ago

    Count me as someone who prefers attaching a couple 6/8 pin plugs to a video card instead of adding the engineering and risk to plowing 300 extra watts across the motherboard.

    • Redundant
    • 3 years ago

    These stories are making the graph line of AMDs Stock look like a heart beat

      • tipoo
      • 3 years ago

      It’s actually been quite the rocket ship since the Zen news. Anyone who bought in a few months ago should be a happy camper.

    • Neutronbeam
    • 3 years ago

    “PCIe 4.0 won’t actually deliver 300 watts from the slot”–I’m not shocked, not shocked to hear that!

    But I got a charge out of the other comments…which seem to be current.

    I’m shutting up and going away now before the ban hammer comes down.

      • K-L-Waster
      • 3 years ago

      This kind of negative post is very polarizing. Try to be more positive, and stay grounded.

        • DrCR
        • 3 years ago

        Yeah, OP, no reason to get short and hot under the collar. It will only amplify things and cause you to burn out.

    • EndlessWaves
    • 3 years ago

    [quote<]Folks hoping for a slot-powered GTX 1180 or Radeon Fury RX will probably have to keep dreaming.[/quote<] Yeah, AMD and nVidia show no signs of bringing power consumption back down to sensible levels.

      • RAGEPRO
      • 3 years ago

      I am very interested in hearing your proposal on how to make a microprocessor with over 2000 functional units and a 256-bit high-speed memory bus slot-powered.

        • Cuhulin
        • 3 years ago

        Magic!

        • Meadows
        • 3 years ago

        Pretty sure you can do it around 300 MHz or so.

          • kmm
          • 3 years ago

          I mean, Snapdragon 820’s Adreno 530 has what, 256 ALUs at over 600 MHz?

    • ronch
    • 3 years ago

    Nah. That GTX 1180 may be happy being just slot-fed, but that Radeon will still need aux power.

    • Meadows
    • 3 years ago

    That’s more like it.

    • Srsly_Bro
    • 3 years ago

    AMD was stoked to hear this. Now it can release Vega and not be out of Spec. Since hearing this news, Vega is delayed.

    • CuttinHobo
    • 3 years ago

    I guess I don’t understand why the spec limits a card to a 300W maximum. Limiting what the [b<]slot[/b<] is rated to supply makes perfect sense, but why limit how much auxiliary power can be supplied? What am I missing? Delivering additional power direct from the power supply shouldn't have anything to do with PCI specifications. 1,000W+ power supplies have been around for ages (and, of course, are total overkill for most of the people who buy them). It should be possible for real halo products to draw more power and put those oversized power supplies to use - with the understanding that appropriate cooling measures be in place.

      • GTVic
      • 3 years ago

      They have to work within overall specs of the systems also. You need a case that can disipate that much heat. Case manufacturers would have to compensate.

      • just brew it!
      • 3 years ago

      They are probably making assumptions about the type of connector used for the auxiliary power connection. Yeah, you could keep pushing things higher by (for example) providing a few SATA style connections to allow people to use extra SATA power leads to help power the GPU.

      At some point it becomes impractical to get rid of the heat, so you’re getting into exotic (and non-standard) card configurations anyway.

        • CuttinHobo
        • 3 years ago

        It does get more and more difficult to dispose of the heat, but I don’t feel that part of it should be PCI-SIG’s responsibility. If there’s demand for a GTX1095 with greater power use, nvidia should be able to make and market a PCI-certified product to users who *aren’t* restricting their gaming box to a SFF case in their entertainment center.

      • kamikaziechameleon
      • 3 years ago

      I think the 300 watt ceiling is a mandate based on practical heat dissipation. If the goal is to really circumvent that limitation those users are so EXTREME that modding the socket in such a way to allow that would not be beyond the relm of possibility for that kind of user.

      Chipset wise we are decreasing wattage pull constantly, generation over generation, and that is a good thing.

        • cynan
        • 3 years ago

        I’m not buying the heat-dissipation argument either. I suppose it a very indirect way heat dissipation is related to distance between PCIe-16 slots insofar as it limits the vertical displacement of an air cooler. However, it has a lot to do with other design aspects of the cooler itself. And also case size and air flow, which is vastly variable and not regulated or explicitly included in some ATX spec as far as I know. And who says halo GPUs need to be stock air-cooled? AMD’s last halo product wasn’t (Fury X).

        Seems like a superfluous spec, that if anything, should be part of the ATX spec, but for some bizarre reason it is covered under PCI Express spec instead.

          • CuttinHobo
          • 3 years ago

          Yes, exactly. This seems more related to ATX specs than PCI.

          Mini-ITX? Maybe that 450W GPU isn’t for you. Good sized tower with sufficient airflow? Buckle up!

    • Waco
    • 3 years ago

    Is it too immature to say “told ya so”?

    Told ya so. 😛

      • DrDominodog51
      • 3 years ago

      No

        • tipoo
        • 3 years ago

        Slightly

    • tipoo
    • 3 years ago

    “Folks hoping for a slot-powered GTX 1180”

    1SM, half disabled?

    • DPete27
    • 3 years ago

    WHAT!? You mean to tell me it’s not possible to deliver 300W of power over 5 microscopic 12V wires?

      • The Egg
      • 3 years ago

      Oh it’s possible, but only if it’s someone accidentally shocking themselves.

      In any other circumstance it won’t work.

      • oba_n
      • 3 years ago

      It’s possible if the voltage is high enough.

        • Meadows
        • 3 years ago

        You mean for extremely high values of 12 V?

          • nexxcat
          • 3 years ago

          12V, 120V, only a single order of magnitude difference. I mean people are [i<]so[/i<] picky.

            • Meadows
            • 3 years ago

            It was a reference to an old nerd joke that went like “2 times 2 equals 5 … for extremely high values of 2”.

            • BurntMyBacon
            • 3 years ago

            If you assume that there is only one significant figure, 2 can be defined as the range [1.5, 2.5). Note the link for those who aren’t familiar with mathematical interval notation:
            [url<]https://en.wikipedia.org/wiki/Bracket_%28mathematics%29[/url<] SQRT(5)=2.236 So it follows that for 26.4% of values accepted for 2, 2 x 2 = 5. Of course if we assume that there is only one significant figure to 5, then 5 can be defined as the range [4.5, 5.5). SQRT(4.5)=2.121 So it follows that for 37.9% of values accepted for 2, 2 x 2 =5, where 5 also only has one significant figure. ... I'll be going now.

            • jihadjoe
            • 3 years ago

            Actually, 54.5V would be enough to push 300W given that the traces are rated for 5.5A.

            • derFunkenstein
            • 3 years ago

            that’s a pretty high value of 12V.

            • chuckula
            • 3 years ago

            I don’t think 54.5V is practical.

            It would take FOREVER for me to click the little plus sign to increment my PCIe voltage in 0.01V steps to go from 12V all the way up to 54.5V. Nobody’s got time for that.

            • derFunkenstein
            • 3 years ago

            Need one of those auto-clicker programs. Do they make those for UEFI interfaces?

            • coolflame57
            • 3 years ago

            Oh come on. At least plagiarize from cookie clicker and get a teleporter to go back in time and get clicks.

        • Eggrenade
        • 3 years ago

        It’s possible if the [i<]current[/i<] is high enough.

          • Waco
          • 3 years ago

          For a very short period of time before the traces vaporize, sure.

            • Eggrenade
            • 3 years ago

            Unless you bought one of the motherboards with the extra pound of copper. Now that purchase is completely justified.

            • Waco
            • 3 years ago

            The connectors are the same size, and they’d melt/burn/flash/stink very quickly. 😛

      • blahsaysblah
      • 3 years ago

      There are 4 reserved pins in 8x slot and 6 in a 16x slot.

      But i can see why they didnt. For a mini-ITX build, you’d still need to take precious motherboard space for 6/8pin(s) and block airflow more.

      They need the ATX spec to be revised with smaller connectors and get rid of all the multi-voltage connectors. Just have simple/tiny hot+ground connectors that you use as you see fit. The new VRM controller on RX480 was able to do even better job w/o a sense pin.

      Its time for a revamp of PS for mini-ITX world and reality of just one adapter in 95% of systems, a GPU. Not the GPU+Sound+Network+HDD cards of a decade ago. They dont update anything and than wonder why PC sales with connectors from decades ago arent a selling point.

      Maybe Cannonlake will bring about a change warranting everyone to upgrade their systems for a real reason for once.

        • UberGerbil
        • 3 years ago

        One of the fundamental design rules for PCIe is that all the power is available no matter how many lanes of data the slot has: the 12V and 3.3V pins are all on the “root” of the slot before the mechanical key, so an x1 has as much power as an x16. They could change that, of course, but I don’t see them doing so.

          • blahsaysblah
          • 3 years ago

          Not true: [url=https://en.wikipedia.org/wiki/PCI_Express#Power<]PCI Express Power from Wiki[/url<] All sizes of ×4 and ×8 PCI Express cards are allowed a maximum power consumption of 25 W. All ×1 cards are initially 10 W; full-height cards may configure themselves as 'high-power' to reach 25 W, while half-height ×1 cards are fixed at 10 W. All sizes of ×16 cards are initially 25 W; like ×1 cards, half-height cards are limited to this number while full-height cards may increase their power after configuration. They can use up to 75 W (3.3 V × 3 A + 12 V × 5.5 A), though the specification demands that the higher-power configuration be used for graphics cards only, while cards of other purposes are to remain at 25 W.

            • Waco
            • 3 years ago

            I wouldn’t trust Wiki on this one, I definitely have a bunch of cards at work that use > 25 watts, yet aren’t a full x16.

            It’s not like the board knows the difference between half and full-height cards either…

            EDIT: Further digging revealed:
            “A x1 standard height, full-length card is limited to a 10W maximum power dissipation at
            initial power up. When the card is configured for high power, by default, it must not
            exceed a 25W maximum power dissipation or optionally it must not exceed a 75W
            maximum power dissipation. A x4/x8 or a x16 standard height or low profile card is limited
            to a 25W maximum power dissipation at initial power up. When a card is configured for
            high power, it must not exceed a 75W maximum power dissipation. ”

            Source: [url<]https://pcisig.com/sites/default/files/specification_documents/ECR_CEM_Power_2009-9-09_2_.pdf[/url<] Basically, anything x1/x4/x8/x16 can run at 75 watts given thermals are taken care of. Note that even for x1 slots, 75 watts is allowed.

            • GTVic
            • 3 years ago

            Seems like it is possible to draw 75W but they don’t want to draw that much power unless it is a 16x graphics card. So that part of the spec isn’t based on capability, they are trying to prioritize less power consumption on smaller cards for whatever reason. Based on comments from AMD, they seem to think that going over on these aspects of the spec is fairly common.

            • Waco
            • 3 years ago

            AMD violated it in a way that was dangerous to motherboards, though.

            EDIT: Downvote all you want, but a card that uses more power than was ever designed to pull through the slot is dangerous, especially with AMD promoting overclocking.

            • danazar
            • 3 years ago

            The board doesn’t know if it’s half-height, but the designer does. Under the spec, the card should initialize at 10W, and then request up to 25W… and it shouldn’t request more than 10W if it’s a half-height card. The card doesn’t have to have AI, it just has to be designed to not request more than 10W if it’s a half-height card.

            Of course the designer can ignore the spec, but then they’re out of spec and can’t promote it as PCIe compliant.

            • Waco
            • 3 years ago

            Absolutely, but at least it wouldn’t damage anything other than the card itself (assuming the card designer didn’t manage thermals properly).

            I don’t care if a card goes over spec in *safe* ways. A half height card with a good cooler won’t have any trouble dissipating 75+ watts. A full size card that uses dual 8 pin connectors with a gargantuan cooling system won’t have issues well beyond 300 watts.

            A card that pulls 25%+ more than designed from the slot? That’s where the card is at fault if anything fails.

            • DPete27
            • 3 years ago

            I see five 12V pins on that diagram.

    • xeridea
    • 3 years ago

    This is good news. 300W max for card is interesting. There have been some dual GPU cards that draw more than 300W, are they just not officially in spec? Does it really matter how much power a card draws as long as it is properly supplied? Lets say you wanted a Dual Titan X running full clocks, which would consume over 500W.

      • RAGEPRO
      • 3 years ago

      I know the R9 295X2 was not allowed to have the PCI Express logo on it anywhere because it was out of spec.

        • odizzido
        • 3 years ago

        Doesn’t mean it’s bad. I don’t know much about this subject but if it’s anything like the WHQL stamp that PCI express logo is worthless.

          • RAGEPRO
          • 3 years ago

          Sure. I didn’t say it was bad. I’d love to have a 295X2. 🙂

      • Topinio
      • 3 years ago

      A dual-GP102 card would most likely be detuned, like the dual-GK110 GeForce GTX Titan Z which was 375 W rather than simply twice the 250 W of the single-GK100 GeForce GTX Titan Black.

      Going over the 300 W specification limit is quite common for dual chip cards: Radeon HD 4870 X2, Radeon HD 6990, GeForce GTX 590, Radeon HD 7990, GeForce GTX Titan Z, Radeon R9 295X2, Radeon Pro Duo all do.

      A couple of dual chip cards (Radeon HD 5970, GeForce GTX 690) were right on the 300 W limit, and it’s mostly the early ones that stayed under it: GeForce 7950 GX2, Radeon HD 3870 X2, GeForce 9800 GX2, Radeon HD 3850 X2, Radeon HD 4850 X2, GeForce GTX 295.

    • chuckula
    • 3 years ago

    Stupid physics getting in the way of all our fun.

      • Redocbew
      • 3 years ago

      I want my flying car!

        • LostCat
        • 3 years ago

        I want your flying car too.

        • TheJack
        • 3 years ago

        You may get one if you wait long enough. I don’t see why Robocopters couldn’t become good enough.

      • ronch
      • 3 years ago

      Well, finances got there way before physics did.

      • tipoo
      • 3 years ago

      [url<]http://i1.kym-cdn.com/entries/icons/original/000/012/978/images_(2).jpg[/url<]

      • Meadows
      • 3 years ago

      See chuckula, that’s how you do the whole fun thing.

    • DancinJack
    • 3 years ago

    300+W from the slot didn’t make sense. I don’t think mobo makers would have liked that, either. Glad someone sorted the info out.

      • Wirko
      • 3 years ago

      On the contrary: they would have a good excuse to introduce new “300 W” motherboards at ridiculous prices.

Pin It on Pinterest

Share This