PCIe 3.0 blamed for Sandy Bridge E delay

Every motherboard maker we talked to at Computex earlier this summer told us to expect X79-based mobos designed for “Sandy Bridge E” processors right about now. I’ll let you in on a little secret: I’m not frantically testing a stack of X79 boards ahead of an imminent launch. Intel’s new high-end platform appears to be delayed, and Charlie Demerjian over at SemiAccurate thinks he knows why. The problem, according to Demerjian’s sources, is the physical interface (PHY) associated with PCI Express 3.0. The PHY is needed to pass PCIe signals outside the CPU, and it’s reportedly only comfortable running at PCIe 2.0 speeds.

Third-generation PCI Express connectivity is built into the Sandy Bridge E CPU, where it will fuel expansion slots in addition to underpinning the DMI interconnect linking the processor to its accompanying X79 chipset. Problems with the associated PHY could not only slow communications with graphics cards, but also cap the bandwidth between the CPU and the chipset. Neither is a particularly appealing prospect for a new flagship platform targeted at servers, workstations, and multi-GPU enthusiasts.

SemiAccurate says the PHY needs to be respun, and the new revision won’t be ready until next year. We could still see Sandy Bridge E debut before then, but it may not do so with its PCIe 3.0 connectivity intact.

Comments closed
    • HisDivineOrder
    • 8 years ago

    Good thing AMD is giving them a lot of breathing room to resolve all their problems. I mean, if AMD could actually get Bulldozer out the gate, they might take advantage of an Intel delay on the high end, but as long as AMD and Intel are both wearing the dunce caps in the corner together, nothing ever changes.

      • NeelyCam
      • 8 years ago

      No – that’s a bad thing. Intel delaying stuff because AMD doesn’t put enough heat on them is a bad thing to everyone but Intel.

    • Bensam123
    • 8 years ago

    This seems remarkably like a rehash of Intels first chipsets running PCI-E 2.0 operating only at PCI-E 1.0 speeds… Remarkably so…

    It is interesting that Intel can have so much trouble implementing a newer version of PCI-E when it’s essentially like prior versions they’ve been working with.

      • willmore
      • 8 years ago

      Actually, PCI-E 3.0 is quite different from previous versions. A quick perusal of the Wikipedia page for it shows that: [url<]https://secure.wikimedia.org/wikipedia/en/wiki/Pci-e[/url<] Not only is the signaling rate increasing to 8 gigatransfers, they're changing the way bits and clock are encoded, upping the equalization, and the enhancing error coding. This is by no means a trivial task.

        • Bensam123
        • 8 years ago

        PCI-E 2.0 had similar changes…

          • Kef
          • 8 years ago

          No, it did not. RTFM…

            • Bensam123
            • 8 years ago

            “The PCI-SIG also said that PCIe 2.0 features improvements to the point-to-point data transfer protocol and its software architecture.”

            [url<]http://en.wikipedia.org/wiki/PCI_Express#PCI_Express_2.0[/url<]

            • willmore
            • 8 years ago

            PCIe was a speed bump (with the same signaling), and some minor improvements to the higher level protocol. Most of the ‘improvements’ were clarafications and other ‘nailing down of lose bits’ from the earlier spec. It wasn’t a completely new PHY design

            • Bensam123
            • 8 years ago

            Indeed, the level of changes isn’t quite the same, but my original point was that Intel had problems going from 1.0 to 2.0 and now is having problems going from 2.0 to 3.0… it seems to be a trend.

            • willmore
            • 8 years ago

            And with that, we agree.

    • Krogoth
    • 8 years ago

    Why does PCIe 3.0 matter?

    The only thing that will use it in the near-future will be high-end video cards which don’t need it.

    The devices (SSD cards, 10Gps ethernet, high-end HBA controller etc) that can use the bandwidth will have to wait a few years before a 3.0 versions get released.

      • Bensam123
      • 8 years ago

      Why does innovation and pushing the envelope matter?

      Consolization, lethargy, and zero drive for accomplishment sucks.

        • Krogoth
        • 8 years ago

        You need killer applications if you want innovation to spring forth.

          • NeelyCam
          • 8 years ago

          You build it, and then they come. Egg/Chicken.

            • Bensam123
            • 8 years ago

            Yes sir…

            …and if people stop building it, less people will be enticed to come to the point at which their is a overall lack of interest in the topic. Even if it seems logical to cut back on R/D because the demand isn’t there at this point in time, it doesn’t mean it will be overly helpful in the long run. Matter of a fact it may function more like slipping the noose over their own head.

            This is completely putting aside trying to be the best you can rather then maintaining the status quo, which generally is a terrible way to go through life.

    • Exo
    • 8 years ago

    Why they called it SemiAccurate? It should be called Totally-NotAccurate…

      • derFunkenstein
      • 8 years ago

      If I had a dollar for every time Charlie was right about something, I could probably get lunch at Panera or something.

        • Goty
        • 8 years ago

        I don’t know, Panera can be kind of expensive…. That Turkey Bacon Bravo is pretty good, though.

          • JustAnEngineer
          • 8 years ago

          +1
          $10 for a sandwich, a bag of chips and a soft drink, but the TBB was tasty.

        • NeelyCam
        • 8 years ago

        Wasn’t that my joke…? Except mine was on Donanim Haber

    • nate011
    • 8 years ago

    It’s hard for me to take any article of Charlie Demerjian’s seriously . . .

    Isn’t he the guy that was always super anti Nvidia and would write ridiculous articles of Nvidia at [url<]http://www.theinquirer.net[/url<] before they kicked him off the website??? (Which is why he is at SemiAccurate now) One of the reasons I quit going to theinquirer.net . . .

      • Farting Bob
      • 8 years ago

      And yet he has a pretty good success rate compared to most rumour sites. Sure its biased, but when the rumours turn out to be correct alot of the time then whats the issue? Ignore all rumours based on “inside sources” and you can be happy.

    • stan4
    • 8 years ago

    Yeah, they won’t release because they don’t have any reason. AMD hasn’t done anything towards their Sandy Bridge chips; They’re just taking advantage and stacking up sales for their previous release.

    • kamikaziechameleon
    • 8 years ago

    I want to read the headline as, “Sandy bridge delayed due to fact that it would force Intel to compete against itself… USB 3.0 is the convenient scapegoated”

    • UberGerbil
    • 8 years ago

    According to [url=http://realworldtech.com/page.cfm?ArticleID=RWT072811020122&p=2<]Kanter[/url<], the PCIe 3.0 speeds are limited to PCIe 3 specifically, not DMI: [quote<]All models have 4 lanes of DMI 2, which is essentially PCI-E 2.0 with certain proprietary extensions. One chip in every system must connect to a southbridge over DMI2, but the others can re-use the interface as a 4-lane PCI-E 2.0 port.[/quote<]And both of those are separate from QuickPath, (aka CSI), which is used for interprocessor communication in the E series. It has been updated to v1.1 with the same 8GT/s transfer rate as PCIe 3.0 -- so it could well have the same PHY issues, but it is a separate protocol that wouldn't get used at all in uniprocessor "Xtreme Nthusiast" X79 implementations. And QPI/CSI is very different from DMI/PCIe, though the SemiAccurate report muddles it all together in their oft-typical "uninformed, confused, or just lazy? You decide!" fashion [quote<]Intel has gone to ‘DMI/CSI/MarketingPablum’ for their interconnect between CPUs and between a CPU and a chipset. [/quote<]

      • phileasfogg
      • 8 years ago

      So, mighty Intel couldn’t get its PCIe3.0 phy shipping before PLX? PLX has about a half dozen PCIe3.0 switches (all with the 87xx prefix) listed on their site – presumably they were/are all waiting for Romley platforms to ship. So, perhaps we will see Intel ship PCIe3.0 on server platforms first, before the technology makes its way to consumer platforms. Sound familiar?? 😉

        • UberGerbil
        • 8 years ago

        We don’t know what the problem is, or even if there is a problem. They may have decided to delay the introduction for other reasons. All we have right now is an apparent change in schedule and a rumor as to its cause.

          • NeelyCam
          • 8 years ago

          Despite his arrogance, biased articles and rants, Charlie (or at least his “sources”) tend to be pretty accurate. He’s breaking stories all the time that turn out to be correct.

          He has some misses too, but I’d give this PCIe rumor a 50:50 chance of being on the money.

            • anotherengineer
            • 8 years ago

            I think the delay was caused by relocating resources to make usb 3.0 look bad and Thunderbolt look awesome 😉

            But I guess they could blame PCIe3.0 also

    • dpaus
    • 8 years ago

    Seems to be a lot of this ‘delayed…’ talk going on; I’m sure it speaks to the renewed sense of competition in the market.

      • crabjokeman
      • 8 years ago

      Actually, it’s probably the lack of competition. Otherwise, they’d be rushing it out the door.

        • OneArmedScissor
        • 8 years ago

        Yeah, screw those thousands of datacenters that would crash and burn.

          • flip-mode
          • 8 years ago

          The two aren’t necessarily mutually guaranteed.

            • OneArmedScissor
            • 8 years ago

            Neither was broken Sandy Bridge #1. Pull that on the server market, and you only have to look as far as AMD’s dwindling market share to see where that lands you.

      • flip-mode
      • 8 years ago

      Hmm… you’d have to tell me more. I don’t see the competitive state of which you speak. Rumors suggest that Bulldozer might not be able to fully compete with current SB offerings, not to mention SB-E, and Bulldozer just got delayed and isn’t likely to be highly available for a few months after launch.

      I find Charlie’s stated reasons for the delay to be believable, though that doesn’t necessarily mean they are fact.

        • drfish
        • 8 years ago

        [quote<]Rumors suggest that Bulldozer might not be able to fully compete with current SB offerings[/quote<] This. As a friend in the know puts it, "AMD will need a billion dollar marketing campaign to sell [i<]one[/i<] Bulldozer chip."

          • kamikaziechameleon
          • 8 years ago

          I doubt bulldozer is so poorly positioned across the board. It will clearly have the upper hand in multi-threaded apps and not in single threaded apps. I think we can all predict that based purely on the performance of AMD’s existing hex core line.

          • PeterD
          • 8 years ago

          But if they save that money on marketing and use it to have lower prices….

      • PeterD
      • 8 years ago

      Economy?

Pin It on Pinterest

Share This