Report: PCIe 4.0 slots will deliver more power and higher speeds

Few peripheral interconnects are as common as PCI Express these days, so major changes to that standard are big news. Last week at IDF, Tom's Hardware got to sit down with the PCI Special Interest Group to get an idea of where the next major update to the standard will take the interconnect technology.

According to Tom's, the next version of the PCIe standard, version 4.0, will have two major updates over PCIe 3.0. For one, each PCIe 4.0 lane will be able to transfer data at speeds of 16 GT/s (just short of 2 GB/s), double that of PCIe 3.0's 8 GT/s. A PCIe 4.0 x16 slot will have a whopping 31.5 GB/s of potential bandwidth on tap, just under twice that of a PCIe 3.0 slot's 15.8 GB/s. Those boosts should give solid-state storage devices and graphics cards plenty of extra headroom to work with.

PCIe 4.0 will also provide much more power to connected devices through the slot itself. Though the PCI SIG didn't offer Tom's a final figure, the vice president of PCI SIG told the site that a PCIe 4.0 slot could provide anywhere from 300W to 400W or 500W, depending on what could be a range of potential specifications. That increase would allow a wide variety of graphics cards, including higher-end models, to draw all of the power they might need though the slot.

The PCI SIG tells Tom's that the PCIe 4.0 spec is on version 0.7 in its lifecycle. Version 1.0 of the spec should be available by the end of this year. Outside of some bleeding-edge products that incorporate the PCIe 4.0 version 0.7 spec today, however, it's not clear when we'll see motherboards or add-in cards that use the new standard for consumer PCs.

Comments closed
    • Waco
    • 3 years ago

    Clarification!

    [url<]http://www.facebook.com/plugins/comment_embed.php?href=https%3A%2F%2Fwww.facebook.com%2FLegitReviews%2Fposts%2F10154059056513882%3Fcomment_id%3D10154059553593882%26reply_comment_id%3D10154064270338882&include_parent=true[/url<] I poked at Legit Reviews, they contacted PCI-SIG to clarify. The slot power limit is *not* changing. They are only certifying additional PCIe connections (6 pin and 8 pin) to allow for > 300 Watt cards.

      • chuckula
      • 3 years ago

      Yeah, the corrections are starting to spread.
      Unfortunately they don’t spread as far & as fast as the initially wrong rumor.

        • Waco
        • 3 years ago

        Unfortunately indeed. I keep seeing people jumping onto peoples cases about how the RX480 must have been designed for PCIe 4.0, blah blah blah. I hate fanboys.

    • dikowexeyu
    • 3 years ago

    They live in a parallel world, out of reality. What a waste of time and money.

    500W is way overkill even for an enthusiast PC with SLI/Crossfire, and the only thing new designs care about, is to reduce power consumption.

    So, the power needs are going down, and they waste time increasing it just to cover rare cases.

    They are pointlessly fighting the last war.

    Also, 2X data speed is too little. It will be rapidly saturated, as SATA 3 was rapidly run over by SSD.

    • shank15217
    • 3 years ago

    Pulling 500W on the pci bus might be dangerous for people who tinker with their boards on an open rack. Better keep those rubber gloves handy…

      • Magic Hate Ball
      • 3 years ago

      Without the high voltage, it’s not super dangerous.

        • Waco
        • 3 years ago

        It’s not dangerous at all, it’s 12 volts.

        • BurntMyBacon
        • 3 years ago

        It is the current, not the voltage, that you need to look out for:
        [url<]http://electrical.about.com/od/electricalsafety/a/amperagekills.htm[/url<] [url<]https://www.physics.ohio-state.edu/~p616/safety/fatal_current.html[/url<] Granted, the human body is fairly resistive (on the order of a megaohm) and electrical currents tend to follow the path of lease resistance. However, when a significantly less resistive path does not exist, the current has to go somewhere. This makes the distinction between current source and voltage source very important. A pure DC voltage source may need on the order of tens to hundreds of thousands of volts to induce the currents necessary to damage the heart or kill you. A pure DC current source will let the voltage go to where the load resistance dictates, but will pass the entire current down the available paths. There are also many environmental considerations like being wet or dirty that can reduce resistance and the human body's AC impedance is not as great either. However, in this case, we are talking about a generally dry (hopefully) clean environment with a DC voltage source of only 12 volts on the PCIe slot. In the end, you are correct in that it is not considered dangerous, but your reasoning needs some qualification.

          • willyolio
          • 3 years ago

          this is a silly argument for people who still haven’t realized that V = IR

          R for the human body is relatively fixed, unless you’re stupid enough to tinker with your computer soaking wet.

          That makes V and I directly related. end of story.

          citing “but omg static electricity has 1000000000000000000000000 billion volts and it only hurts a bit” completely fails to account for total [b<]energy[/b<] while pretending that V and I aren't directly tied to each other. arguments like that only demonstrate complete lack of understanding of how electricity works.

            • Beltonius
            • 3 years ago

            Yea, but it only takes 5 or more milliamps across your heart to potentially cause life threatening danger (this is why GFCI outlets open up when there’s 5mA of leakage from line to ground)

            At 500W, there would be over 40A available. 150mA of DC current is enough to cause your heart to misfire, according to the NIH: [url<]http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2763825/[/url<] 300mA of DC current through your arm can be enough that you are unable to let go of something giving you that shock. Your resistance between oppposite hands (that is, the resistance across your heart) can vary by several orders of magnitude. Generally it's in the 10-100 kOhm range. However, it varies by person, and its not unreasonable for a sweaty hand to be in the sub 1kOhm range. The NEC (National Electric Code) and NFPA (National Fire Protection Agency) 'Class 2' power supplies are considered generally safe, and they limit current to 8 A below 20 V, and a max power rating of 5 times the nominal voltage (60V). These limits are chosen because they are generally considered safe, both for general human contact as well as preventing fires.

            • Waco
            • 3 years ago

            Car batteries have hundreds of amps at 12V available. You’d have to arrange very very specific circumstances to shock yourself with your hands alone, even with terminals very close together.

            Hell, I have a 5 Farad 12V cap in my car. With any small amount of common sense (no super salty wet hands while touching each side of the cap with one finger on each hand), it’s not dangerous.

    • EndlessWaves
    • 3 years ago

    Someone told the PCI-Express SIG about global warming.

    They thought it sounded like a good idea.

    • Pax-UX
    • 3 years ago

    The future of case wiring and airflow sound good. NVMe ssd will be into slots leaving almost nothing needing extra cables, some USB cases port and the standard case lights & power. As someone looking to build a MTX; losing all but 2 SATA ports would be nice since those would only really be use to bulk storage devices 1TB SSD or a nice 4TB HD. But even those now I would sooner pick up as USB3 or NAS.

    • JosiahBradley
    • 3 years ago

    For those of you wondering why the PCIe spec needs beefy power delivery, don’t forgot about external enclosures and remote PCIe. They’re doing this to match/beat parity with thunderbolt. Also no one here remembers overclocking the PCIe bus using dedicated floppy PSU headers?

      • ImSpartacus
      • 3 years ago

      I totally forgot about external enclosures. Good catch.

      That makes a hell of a lot more sense.

        • Krogoth
        • 3 years ago

        It is also for daughterboards on high-end server, datacenter and HPC boxes.

          • ImSpartacus
          • 3 years ago

          I’m not familiar with those, but couldn’t they just use an external peg connector like a graphics card?

          However, you can’t do that with an external pcie connector (for an external gpu or whatever) because then you’d need to also cram in an external psu and external power connector to the wall (like the razer core). Pushing more power through pcie makes external graphics much simpler (and potentially cheaper). That feels like the “killer app” for going through the trouble to get several hundred watts of power through the poor motherboard.

          But maybe there’s some sever/hpc-specific quirk that makes high power pcie useful.

            • Krogoth
            • 3 years ago

            They are found more in the realm of custom-hardware platforms in HPC and high-end datacenter boxes where the designers are trying to cram as much memory, CPU, GPU power as possible within thermal limits.

    • HisDivineOrder
    • 3 years ago

    And this will change nothing about power delivery on expensive, high end GPU’s. At least, not for a long, long while.

    Of course, at first, it’ll be about backward compatibility with older slots. Eventually, it’ll be about perception versus reality. You know the gamers that buy a 1080 for 1080p and don’t downscale or do anything or even have a monitor capable of doing faster than 59 fps?

    Those are the ones that’ll DEMAND that TWO eight pin power slots get on any card they’ll consider. It won’t matter how many times we say, “But, dude, the spec has enough power for it!”

    They’ll scream, “I want HIGH END hardware. Not low end. HIGH END!”

    • TwoEars
    • 3 years ago

    So now I have an update plan for my 4790k system… whenever PCIe 4.0 comes out. Good stuff.

      • cynan
      • 3 years ago

      It’s telling of the relative stagnation of CPU performance that PCIe versions are a more motivating reason to upgrade your system than to have a platform to support the new performance hotness of a new CPU. On an enthusiast forum.

      And I’ll probably try to do likewise. My main system is still rocking Sandybridge. Wouldn’t you know it, the first consumer chipset to support PCIe 3.0 (X79)

    • RdVi
    • 3 years ago

    The end of this year? So it will launch just in time to make Zen have an outdated chipset… or in this case, SoC?

      • albundy
      • 3 years ago

      beat me to it! and yes.

      • anotherengineer
      • 3 years ago

      DP 1.4 was released March 2016 and DP 1.3 in Sept 2014, so all monitors are way outdated………….unfortunately……………..

    • Meadows
    • 3 years ago

    Up to 500 W delivered through the motherboard itself? Won’t that need some sort of extra insulation or significant extra weight in components to work? I’m no engineer but it sounds like an unsafe jump. Unnecessary, too.

    I hope AMD has eyes and ears at this special interest group, otherwise they’re off to a pretty bad start with Zen, should their AM4 motherboards be behind the times again.
    For reference, AM3 boards never got to support PCIe 3.0 and while the FM2 boards did, you couldn’t buy high-performance CPUs for those. (As “high” as AMD’s go, anyway.)

    It would be nice if they could manage a product launch that doesn’t need anything replaced within a year.

      • chuckula
      • 3 years ago

      AMD will just implement PCIe 4.0 in the next iteration of Zen in 2018 or so.
      It’s highly unlikely that there will be a consumer-level platform with PCIe 4.0 out in 2017 from any manufacturer, so Zen isn’t being hurt by this any more than Skylake E will be hurt since it will use 3.0 as well.

      • Krogoth
      • 3 years ago

      It is unlikely you see that kind of power usage for PCIe perpherials in non-enterprise tier hardware. It is for external PCIe enclosures and daughterboards.

      The motherboards that going to carry that level of current are going to be non-ATX tier stuff.

        • End User
        • 3 years ago

        [quote<]It is unlikely you see that kind of power usage for PCIe perpherials in non-enterprise tier hardware[/quote<] Dual OC'ed Volta based Titans in SLI?

    • ronch
    • 3 years ago

    Get a move on, AMD. Call Asmedia now.

    • Mikael33
    • 3 years ago

    I imagine AMD must be happy about this.

      • chuckula
      • 3 years ago

      They were right.
      The Rx 480 was future proof.

        • tipoo
        • 3 years ago

        Even though it wasn’t built for PCI-E 4, I wonder if they could detect when it’s plugged into a system with it and revert to the old power mechanism…

        • ronch
        • 3 years ago

        And today’s Sarcasm Awardยฎ goes to… CHUCKULA!!!

        • Mikael33
        • 3 years ago

        Classic AMD over engineering their Gpus for the future.

          • tipoo
          • 3 years ago

          You’ll all see, DX13 will make Polaris blow Pascal away!

        • ImSpartacus
        • 3 years ago

        I lost it.

      • Goty
      • 3 years ago

      Came here for this joke. Was not disappointed.

      • maxxcool
      • 3 years ago

      Very nice … ๐Ÿ™‚

    • chuckula
    • 3 years ago

    [quote<]Though the PCI SIG didn't offer Tom's a final figure, the vice president of PCI SIG told the site that a PCIe 4.0 slot could provide anywhere from 300W to 400W or 500W, depending on what could be a range of potential specifications.[/quote<] If I were a mobo manufacturer I'd be strongly lobbying for just the opposite. You'd probably have to support the existing power specifications for backwards compatibility, but I'd push as hard as I could to have any high-power device (GPU) be supplied with power direct from the PSU as much as is possible to cut the material & engineering costs of the motherboard.

      • Waco
      • 3 years ago

      I’m imagining this is, again, a total misrepresentation of the spec. Just like the AMD fanboys that couldn’t read the spec carefully, I’m expecting that the slot will still be limited to 75 watts or so, and that the *specification* will support additional PCIe power connectors up to higher limits that previously allowed (e.g., dual 8 pin + slot, triple 8 pin, etc).

        • chuckula
        • 3 years ago

        I could buy that interpretation.

        I remember that doc about the maximum allowable power draw being grossly misinterpreted to mean that the max allowable power draw was coming solely through a dinky PCIe slot when that clearly was not true (and if it were true then 6 and 8 pin PCIe cables would literally have no reason to exist).

      • arunphilip
      • 3 years ago

      Came here to say just this – if I were a mobo designer, I’d dread the additional and thicker traces required for that 4x higher wattage.

      When CPUs are settling around the ~100 W mark, one wonders why the mobo has to be the component in charge of delivering 300 W power to a single component, instead of a dedicated line from the PSU, as is done today.

      • synthtel2
      • 3 years ago

      I too came here to say just this. ~200 watts (CPU OCing) takes pretty fat traces and a nearby plug. ~2.5x that current sounds like something mobo people would really really not want to deal with, something that would take its own nearby connector on the mobo a la EPS (kind of negating the point), and something that would require some serious revision to the slot physical spec (probably breaking backward compatibility).

    • chuckula
    • 3 years ago

    [quote<] A PCIe 4.0 x16 slot will have a whopping 31.5 GB/s of potential bandwidth on tap, [/quote<] Yeah, but we all know Ngreedia will advertise it as 32 GB/sec and not tell anyone about the other 0.5!

      • Redocbew
      • 3 years ago

      A dead horse is dead.

        • chuckula
        • 3 years ago

        A classic never goes out of style!

          • Concupiscence
          • 3 years ago

          It really, really doesn’t.

          • derFunkenstein
          • 3 years ago

          That’s true, but in this case it doesn’t apply. Nothing classic. ๐Ÿ˜‰

      • Srsly_Bro
      • 3 years ago

      1/10

      Stop trying so hard, chuck. That was…reaching a bit.

    • AnotherReader
    • 3 years ago

    So, is it safe to assume that we won’t see this in any platforms until mid 2018? PCIE 3.0 was finalized in [url=http://www.xbitlabs.com/news/other/display/20101118151837_PCI_Special_Interest_Group_Publishes_PCI_Express_3_0_Standard.html<]November of 2010[/url<], but [url=http://www.anandtech.com/show/5771/the-intel-ivy-bridge-core-i7-3770k-review/5<]no consumer products supported it before the launch of Ivy Bridge in April 2012[/url<].

      • chuckula
      • 3 years ago

      Very true. There’s always a non-trivial lag between the final standard being ratified and products being sold.

      • Redocbew
      • 3 years ago

      PCIe 4.0 isn’t finalized yet, so products using it may be even farther out than that.

      • cynan
      • 3 years ago

      X79 could be considered a consumer product.

    • Chrispy_
    • 3 years ago

    The 24-pin ATX connector spec doesn’t currently handle 300-500W though. That’s why most boards need a dedicated 4-pin or 8-pin connector for the CPU (each pair of +12V circuits provides 75W)

    What the PCI SIG are proposing means a revision of the ATX standard too, meaning every power supply that doesn’t have a new PCIe 300+ connector for the motherboard on it will require adapters.

    Of the 24 pins in the standard ATX adapter, only two are +12V and the rest are 3.3 or 5V circuits and ground/sense/signal wires. The most sensible option is to allow 8-pin PCIe power plugs to connect direct to the motherboard, which means a whole bunch of extra traces on the motherboard just to carry power that otherwise goes directly to the GPU in the first place.

    As long as every PSU in the land doesn’t suddenly need replacing or a whole bunch of expensive adapters with silly markup, I’ll leave my pitchfork in the shed.

      • xeridea
      • 3 years ago

      Video cards would also be more complex, with need to be able to pull power from either source, and there would be extra confusion from users needing to check what their board supported, or where to plug in the connector, since it is not possible to get this power from the ATX plug. Seems like a pointless change to me.

      Extra speed would be useful for some niche applications, but GPUs aren’t bottlenecked from the PCIE bus.

      • Redocbew
      • 3 years ago

      That was the first thing I thought of also. If they want to create a slot which can take that much power that’s great, but there’s got to be a way for that much power to get to the slot in the first place.

      • Krogoth
      • 3 years ago

      The power requirements are going to be a non-issue for non-enterprise tier hardware. PCI SIG is just covering the bases for their enterprise clients with PCIe 4.0 spec.

        • ImSpartacus
        • 3 years ago

        Can you elaborate on why they wouldn’t be an issue for non enterprise use cases?

          • Krogoth
          • 3 years ago

          It is because mainstream-tier PCIe peripherals are not going to use anywhere need that amount of current. The current trend in this market has been reducing power consumption and increasing energy efficiency. Crazy SLI/CF on a single-card solutions barely hit ~300W and thermal management for such cards is a PITA.

          Motherboard guys are going keep making boards spec to the PCIe 1.0 as the baseline for power delivery. The only thing that PCIe 4.0 support will offer for non-enterprise users is increased bandwidth.

            • Waco
            • 3 years ago

            The slot power isn’t changing, the card power is. The more I read about this the more I think everyone is misinterpreting the spec just like all of the AMD fanboys who wanted to justify why the RX480 wasn’t dangerous.

            The slot power, I’d bet, only moves up to the 100 W range at most. The rest of the power will come from more certified combinations of PCIe power cables…

      • dikowexeyu
      • 3 years ago

      So, If I want a new PC, I will need to replace my expensive 750W PSU with a 250W PSU (which will be overkill even for powerful PC’s), just to get the new connector, and a marginally faster computer.

      That’s just another reason to avoid upgrading.

      The only constant on AMD/Intel/nvidia/storage manufacturers/etc is to show how much less power his stuff will consume in the future.

      • psuedonymous
      • 3 years ago

      The ‘lazy’ way to implement that would be to add PEG power plugs to the motherboard. That way the ATX 2.4 standard itself does not need to be revised, and maintains backward-compatibility.

    • tipoo
    • 3 years ago

    PCI-E 4: Now officially supporting Overwatts.

    • ronch
    • 3 years ago

    Your days are numbered, 6-pin aux power connector.

      • Airmantharp
      • 3 years ago

      Oh, they’ll still exist- they’ll just be plugged into the board instead.

      You know, until the ATX connector to the board gets an update.

      • Chrispy_
      • 3 years ago

      Nah, you’ll either need to throw out your PSU too, or more likely just plug the 6 or 8-pin aux cable directly into the motherboard instead.

      Your existing PSU isn’t going to magically develop 6x more 12V rails on the 24-pin overnight you know ๐Ÿ˜‰

      • UberGerbil
      • 3 years ago

      Don’t count on it. The SIG is getting some pushback from the mobo folks, who aren’t eager to put that much power through the motherboard (and all the layers and isolation that requires). There’s likely to be some “levels” or “optional features” in the power-delivery side of the spec, allowing the mobo makers to offer cheaper motherboards that provide max PCIe 4 [i<]bandwidth[/i<] but not max [i<]power[/i<], reserving full-power for the higher-end, more expensive motherboards. That means that graphics cards will have to retain the connector for the situations where they're paired with a cheaper motherboard. And then there will be the GPUs that require both 500W from the motherboard [i<]and[/i<] 250W from auxiliary connectors for MOAR FRAMERATEZ!

    • DancinJack
    • 3 years ago

    [quote<]A PCIe 4.0 x16 slot will have a whopping 31.5 GB/s of potential bandwidth on tap, more than twice that of a PCIe 3.0 slot's 15.8 GB/s.[/quote<] Say what now?

      • Jeff Kampman
      • 3 years ago

      Whoops, fixed.

        • DancinJack
        • 3 years ago

        ๐Ÿ™‚ Thanks!

    • Redundant
    • 3 years ago

    500W? Nice, AMD can make a GTX1080 equivalent

      • tay
      • 3 years ago

      [url<]http://67.media.tumblr.com/tumblr_m0nmin7ast1qibz0jo1_500.png[/url<]

      • Srsly_Bro
      • 3 years ago

      I found that funny. I’m not a cry baby tho.

      • DancinJack
      • 3 years ago

      Pretty funny.

      • tipoo
      • 3 years ago

      Mental image of an AMD exec furiously scribbling down notes

Pin It on Pinterest

Share This