Final DDR4 specification published

Memory makers have been showing off DDR4 memory modules for months. The final specification covering the replacement for DDR3 was only finalized today, though. The JEDEC group that oversees memory standards has published the DDR4 spec, which can be downloaded here. Registration is required to grab the 214-page PDF, but the highlights can be seen on this page of JEDEC’s site. Here’s a taste:

The per-pin data rate for DDR4 is specified as 1.6 giga transfers per second to an initial maximum objective of 3.2 giga transfers per second. With DDR3 exceeding its original targeted performance of 1.6 GT/s, it is likely that higher performance speed grades will be added in a future DDR4 update. Other DDR4 attributes tightly intertwined with the planned speed grades, enabling device functionality as well as application adoption, include: a pseudo open drain interface on the DQ bus, a geardown mode for 2,667 MT/s per DQ and beyond, bank group architecture, internally generated VrefDQ and improved training modes.
The DDR4 architecture is an 8n prefetch with two or four selectable bank groups. This design will permit the DDR4 memory devices to have separate activation, read, write or refresh operations underway in each unique bank group. This concept will also improve overall memory efficiency and bandwidth, especially when small memory granularities are used.

Although the specification has been finalized, we probably won’t see DDR4 make a dent in the DRAM market for some time. iSuppli expects DDR4 to make up only 12% of the module market in 2014—and none of it before then. In 2015, the firm anticipates over half of the memory modules sold will have DDR4 chips onboard.

New memory types tend to be comparatively expensive to start, so a slow transition to DDR4 isn’t a bad thing. Existing DDR3 memory seems plenty quick for desktop systems, and it’s affordable enough that 16GB configurations can be had for less than $70 right now.

Comments closed
    • Antimatter
    • 7 years ago

    If my understanding is correct DDR3 uses 64 bit channel and in dual channel you get an effective 128 bit bus. Assuming there are not cost and power penalties why not double the channel width to 128 bit with DDR4. That alone would double the bandwidth on top of any clockspeed increases.

      • jihadjoe
      • 7 years ago

      There are cost issues, and making a gangable narrow bus is the solution.

      Premium platforms can go dual, triple or quad channel, meanwhile smaller platforms like SOCs can run a cheaper single channel implementation.

      Even GPUs are multi-channel, but advertised as a wider bus. i.e., Nvidia’s GTX680/670 actually use 4 64-bit wide interfaces to memory (quad channel). In the 660Ti this is cut down to three (triple channel), but in the advertising literature they say it’s 256bit, or 192bit. A quad channel desktop like X79 is effectively a 256-bit interface to DDR3.

    • Alexko
    • 7 years ago

    “Existing DDR3 memory seems plenty quick for desktop systems”

    With current—and especially upcoming—APUs, that’s very debatable. I for one think DDR4 can’t come soon enough.

    There’s a reason Intel is putting DRAM on the CPU package for Haswell.

      • Airmantharp
      • 7 years ago

      More bandwidth will help APUs, but they’re still a comparatively narrow niche. Putting DRAM on the CPU package will create a performance advantage that even moving to DDR4 wouldn’t grant Intel.

      Mostly, I agree that there’s very little reason to push for DDR4’s adoption. I’d rather DDR3 stay cheap until consumer CPUs make make use of the 2400MHz stuff that seems so easy to make.

      • bcronce
      • 7 years ago

      I’m less excited about the increased bandwidth than the much lower power consumption.

    • zzz
    • 7 years ago

    With companies pushing out next-gen products, USB 3, DDR4 and HTML 5, why are these groups so slow to push out a standard? Companies are pushing out products years before they decide on a standard: why is it taking them so long? Either stop posting white-papers or formally create standard protocols.

    Cause really, Asus’s revamp of USB 3 should be be ‘normal’ usb 3, but it’s not.

      • BIF
      • 7 years ago

      I believe the slowness is because any standards organization is mostly political.

      This goes for building codes, cosmetology licenses, and even your local homeowner association landscaping approval committee.

    • just brew it!
    • 7 years ago

    Something people tend to ignore is that while burst transfer speeds have increased steadily as we’ve progressed from SDR -> DDR -> DDR2 -> DDR3, latency (in absolute terms) has changed little in the past 15 years. Larger caches and multi-channel memory interfaces partially mitigate this, but faster transfer speeds are a matter of diminishing returns unless you’re running code that sequentially streams large amounts of data.

      • Wirko
      • 7 years ago

      [url=http://www.old-computers.com/history/detail.asp?n=43<]This tame looking hexadecipede[/url<] had an acces time "between 100 and 150 nsec maximum" in 1979.

        • just brew it!
        • 7 years ago

        Yup… and by the late 1990s access times had improved to around 10ns. We’ve been stuck in that general vicinity ever since, while in the meantime burst transfer rates have increased by something like a factor of 25.

          • Wirko
          • 7 years ago

          The delay (now called latency) in those old DRAMs was specified for random access if I recall correctly. A 4 MHz Z80 processor, for example, would work with a 200-250 ns DRAM.
          In DDR3, the latency for completely random acces is around 35 ns (calculated from tRAS+tRP instead of tCAS). Not so good for a”random access” memory, huh?
          So a worst case comparison, though it’s not really meaningful, would give a factor of 3 or 4 in 33 years.

      • bcronce
      • 7 years ago

      I hope memresistors help reduce latency be removing pre-charge latencies completely.

    • Grigory
    • 7 years ago

    In 1990 east Germany upgraded from DDR to freedom.

      • NeelyCam
      • 7 years ago

      In 2001 USA downgraded from freedom to fascism

        • Grigory
        • 7 years ago

        Going for the Big Lie or is it projection? Your ilk must have been very sad in 1990.

          • NeelyCam
          • 7 years ago

          Some people call it the Patriot Act

            • Grigory
            • 7 years ago

            Sorry, I spent my youth in a country sized socialist Gulag (not the DDR, tho) so this is touchy subject for me. Please, don’t compare something like the Patriot Act (extended by Obama last year btw) with fascism, something I have first hand experience with in the red flavor. Thank you.

            • Jing
            • 7 years ago

            Why not? I too came from a former communist/socialist country, and I have no trouble seeing the Patriot Act as fascist. If anything, it’s more insidious because most people would not be able to recognize it as such. And it will be too late to reverse the trend when they do, if ever.

            • no51
            • 7 years ago

            You forget that America is #1 at complaining and hyperbole.

            • DrCR
            • 7 years ago

            You’re right. We should trust our politicians.

        • Buzzard44
        • 7 years ago

        Now why do you have to take a perfectly good joke and use it to troll with?

    • flip-mode
    • 7 years ago

    One day I’ll upgrade my home desktop from the DDR2

      • tfp
      • 7 years ago

      The same day you upgrade your home desktop?

    • ronch
    • 7 years ago

    It’s good to see PC technology moving forward, but in another article here today, mobo makers foresee slumping demand. Many folks just find their systems adequate and can’t see any reason to upgrade. Still, it’s nice to have much better PC parts available for when my system eventually bites the dust and I have to go out and buy new parts.

    • sschaem
    • 7 years ago

    It would have taken way over a decade to go from DDR2 to DDR4.

    I wonder if rambus is the reason the memory research as been near stagnant?

    HMC cant come soon enough…

      • chuckula
      • 7 years ago

      Rambus, while annoying, is not why memory standards have slowed down. The real issue is that for a large number of devices, DDR3 is not really a huge bottleneck right now. Will DDR3 be a bottleneck in the future? Sure, but by then DDR4 will be established.

      Memory is not developed in the same way that CPUs are, where the “standards” are mostly set by one or two big manufacturers. Instead, all of the RAM companies get together and painstakingly debate and deliberate over new standards. It is an annoying and slow process, but it means you can get highly compatible chips from a bunch of different manufacturers, which is why RAM prices are as low as they are.

        • UberGerbil
        • 7 years ago

        Exactly. And this is the reboot of DDR4, because the last attempt atan agreement collapsed and they had to go back to the drawing board.

        The other factor in the relatively long reign of DDR3 are the rather unique macro-economic climate of the past few years: the global economic downturn arrived just about at the point where many of the RAM players were gearing up for the next generation, and they all pulled their heads in (the NAND guys got trapped with oversupply and had to sell below cost for a while). The whole business model for the memory manufacturers involves enormous capital outlays initially to get the new lines up and running; they then try to earn that back (and some profit) though high initial prices. That works until enough other entrants have driven prices down to the point where they’re basically selling at little more than cost (and only the most efficient players with the largest volumes have any advantage). The result is that the transition to a new flavor of RAM is an enormous, multi-billion dollar game of chicken, where the first entrants are most likely to make a profit but also incur the highest costs (since they’re the first customers for equipment and have to work out any kinks), and also face the weakest demand. Which is why they’re always trying to do things like ride Intel’s coat-tails, arrange illegal cartels, and sue one another.

        • sschaem
        • 7 years ago

        DDR3 on a quad core APU seem like a bottleneck.
        Its also an issue with many core processor, that why Intel produces quad channel ddr3 controlers.
        No need to go with that expenses if DDR3 was adequate.

        And Rambus as no hands in holding memory technology IP over the committees heads?

        To me DDR4 is just too small of a step forward, way to late.
        I personally not going to invest and support it.

        I think only massive leap like HMC will move the computing industry forward, not slow and iterative upgrade like DDR4.

      • NeelyCam
      • 7 years ago

      [quote<]HMC cant come soon enough...[/quote<] Couldn't agree more

    • anotherengineer
    • 7 years ago

    Where’s the shortbread? Is it going to be a late evening diner roll again?

    • chuckula
    • 7 years ago

    IT’S OVER
    — Strong Bad

      • DrCR
      • 7 years ago

      You’re totally loosing your edge.

    • adisor19
    • 7 years ago

    And now to address the elephant in the room : will Rambus get to charge royalties for the memory makers that choose to fab DDR4 RAM ?

    Adi

      • flip-mode
      • 7 years ago

      And will Apple somehow claim Rambus has violated 1,337 of its patents?

        • Bauxite
        • 7 years ago

        1 patent troll deserves another.

        • Kiwi386
        • 7 years ago

        They better not make those DIMMs rectangular with rounded corners ……..

      • NeelyCam
      • 7 years ago

      Probably. After all, they are one of the premium interconnect research institutions – they most certainly have some important patents on some of this stuff.

        • UberGerbil
        • 7 years ago

        Rambus really does have some legit patents on arguably clever innovations; despite their legal behavior, they are (or at least were once) a real technology company that produced genuine intellectual property (rather than just buying patents for trollish purposes). Certainly they’re no worse than the other memory manufacturers (as the courts eventually recognized), though they got tarred in enthusiasts’ minds early and never overcame that. Which is not to say I find them particularly admirable, but they certainly don’t deserve to be singled out to the degree they usually are.

          • Airmantharp
          • 7 years ago

          They got tarred by getting Intel to push their product that performed worse than the existing solution at around eight times the cost. I remember that crap. RDRAM had no place being hooked up to Pentium IIIs.

            • UberGerbil
            • 7 years ago

            Sure, no argument. But the opprobrium there should be directed at Intel. Intel should have known better than to tie its product to a memory tech that didn’t have wide support in the industry (and was unlikely to get it due to Rambus stance on license fees), even if they really thought that was the only way to salvage the performance of their looming speedracer PIV design.

      • Buzzard44
      • 7 years ago

      Whoa, DDR4 has elephant addresses?!

        • DrCR
        • 7 years ago

        I think Adi meant DDR4 now has bespoke pointers for any elephants that may be present.

      • danny e.
      • 7 years ago

      Rambus should probably sue Apple and ask the judge to have Apple cease and desist from selling or marketing any and all of their products.. since all of their products use “memory” which is obviously an invention of Rambus.

      Sadly, even Rambus will die when Al Gore sues every electronics maker for using the “internet” which he clearly invented and owns.

Pin It on Pinterest

Share This