Back to Article

  • richardginn - Saturday, August 02, 2014 - link

    Newegg,.com has some up as well:
  • Devo2007 - Saturday, August 02, 2014 - link

    Interesting all theirs are labeled "server memory" (and in stock too!) I did get a laugh out of the related items showing DDR3 boards since DDR4 ones aren't available yet. Reply
  • HanzNFranzen - Saturday, August 02, 2014 - link

    but those are all ECC memory modules, not the non-ECC desktop parts. Reply
  • HanzNFranzen - Saturday, August 02, 2014 - link

    2133 C16... ouch... that's a pretty significant drop in performance compared to the DDR3 equivalent, is it not? I realize it is lower voltage, but is there something else about DDR4 tech that will make up for the difference in higher timings? Or is this a matter of waiting it out for R&D to bring the timings back down? I'm anxiously awaiting X99 for my next upgrade, and I understand DDR4 is new and DDR3 is mature, but didn't think I would be taking such a step back as far as memory performance. Reply
  • ImSpartacus - Saturday, August 02, 2014 - link

    Isn't there always some reason why timings seem to go up but it's never a big deal? Reply
  • tipoo - Sunday, August 03, 2014 - link

    Not at the same clock speed. But often, a higher latency in cycles is offset by a higher clock speed, making the time interval (which is what matters) similar. That's with higher speed modules of the next DDR generation though. Reply
  • StevoLincolnite - Wednesday, August 06, 2014 - link

    It's latency per clock cycle.

    Lets put it this way, If you theoretically had DRAM running at 1ghz with a CL10, then that will have the same latency as DRAM running at 2ghz with CL20.
  • modulusshift - Saturday, August 02, 2014 - link

    Um. What are your current timings? Current DDR3 standard timings seem to be 1333 C9 and 1600 C11. High end is marketed as 2133 C9 and 2400 C11.
    Considering 2133 C15 is the low end, anything above 2400 C14 is pure gravy. (remember C-latency is calculated in terms of the clock speed at half the data rate. So it's only comparable between RAM at the same speed otherwise.)
  • trichome333 - Sunday, August 03, 2014 - link

    Never run 1333 @ C9. I think it was on here they showed that all the mainstream timings were within 1-2% of each other performance wise besides DDR3 @ 1333mhz and CL9. I run 1333 @ CL8 and with everything else the same (besides the other timings of course) Im actually able to get slightly higher scores in Sandra and PC Mark because my sticks run at 1T instead of 2T. I see a lot of people toss up benchmarks and instantly can tell they havent set their timings. I have a basic understanding of what most of them do but I am great at figuring out the best timings just by the feel of the machine. Definitely not the sexiest part to know but once you get good with timings youll realize basically every stick has a sweet spot (unless its defective).
    On another note, remember why GPU's skipped over DDR4 besides a few ATi cards? Because its not much better than DDR3. 3k at C15? Maybe. But all the others look pretty comparable to a nice 1866 kit. We shall see.....
  • jabber - Monday, August 04, 2014 - link

    Generally I just find the timings for the next lower speed (basically run 1600 with 1333 timings etc.) and apply those. Works pretty well most of the time. Reply
  • frenchy_2001 - Monday, August 04, 2014 - link

    Graphic cards have not skipped anything. They use DDR3 (for cheap ones) or GDDR5 (which is a graphic centric evolution of DDR3. Higher bandwidth with higher latency).
    I am sure the cheap cards will switch to DDR4 when it makes sense cost-wise, while they may develop higher graphic memory based on DDR4 if needed.
  • DanNeely - Sunday, August 03, 2014 - link

    Look at the faster parts; 2400 is also C16, while 2666 and 3000 and C15. The 2133 and 2400 are clearly dump parts for low binned chips (which shouldn't be surprising since they're DDR3 1066/1200 equivalents). DDR4 3000 C15 is the rough equivalent to DDR3 1500 C7.5 which is much better than the timings modulusshift is talking about; and in terms of latency in ns corresponds to an equivalent of 2250/ C11.25. Reply
  • Impulses - Sunday, August 03, 2014 - link

    To think I bought my current 1600 4x4GB DDR3 kit for like $60, and that was a year and a half ago. Reply
  • HisDivineOrder - Sunday, August 03, 2014 - link

    Them were the days, weren't they?

    That's the ultimate problem with the value proposition of Haswell-E. They finally get around to making it so a hexacore CPU is going to be priced at a halfway decent, easily justifiable price point and then they build it around a memory tech that will make up for every bit of that price difference.

    So close, so far away.
  • ZeDestructor - Sunday, August 03, 2014 - link

    Prices are in-line with DDR3. Welcome to UK "tech tax":

    For an actual comparison, using that exact site, the 4x8GB DDR3-2400 CL11 Corsair kit costs £339.95 ( (with the same kit on costing $359.99 ( compared to the 4x8GB DDR4-2400 CL16 Crucial kit which costs £359.99 (

    If you ask me, £20.04 difference is completely acceptable for brand new tech with what will be extremely low demand from consumers because it will be X99/LGA2011-3 only, and LGA2011 Core i7 users are fairly rare, with most LGA2011 users using Xeons with ECC memory and not caring about prices.

    I'm slightly surprised at CL15-16 though.. I expected CL12-14 personally...
  • trichome333 - Sunday, August 03, 2014 - link

    I got an 8gb kit of Crucial Ballistix free with an $80 Biostar mobo about that long ago too. Both have been solid. Although I remember getting a summer job to get an extra 4mb for my rig to play DOOM II =) Before that I couldnt load Windows because it would leave me short on RAM lol. Had to load her up in DOS. Reply
  • Kevin G - Monday, August 04, 2014 - link

    Aye. I remember picking up some 16 GB registered ECC modules around the same time frame for $125 USD per DIMM. Memory was cheap back then. Reply
  • R3MF - Sunday, August 03, 2014 - link

    wake me up when we get to 3200 speeds for ~£8/GB Reply
  • haukionkannel - Sunday, August 03, 2014 - link

    Sleep well ;-)

    It will take some time. If we look back when ddr3 was released these seems to be "reasonable" priced compared to that time. Of course after that there was fierce competition that reduced the price of DDR3 to quite low. Not going to happen this time, when considering the price of DD3 at this moment. So few makers in the market these days...
  • StrangerGuy - Sunday, August 03, 2014 - link

    There highly unlikely that there will be yet another record low <$100 for 16GB DDR4 like the case in DDR3, unfortunately. Desktop/laptops that uses traditional DIMMs are in decline, the number of DRAM manufacturers have shrunk, plus they would rather cater to LPDDR and NAND production which are much more in demand. Lastly, they won't be gullible enough again to overproduce DDR just because there is another new version of Windows around the corner. Reply
  • mapesdhs - Sunday, August 03, 2014 - link

    IMO more important for X99 that speeds is capacity. If X99 is stuck with the same 64GB
    limit as X79, then a lot of potential solo professionals won't bother with it because 64GB
    just isn't enough these days, not when imaging/video demands are starting to move up
    to 4K. AE users inparticular need a big step up the scale, preferably at least 256GB max.

  • mapesdhs - Sunday, August 03, 2014 - link

    (sorry for the typos... really, we still can't edit our posts??...) Reply
  • Kevin G - Monday, August 04, 2014 - link

    Except that DDR4 will support much larger module sizes than DDR3 down the road. 16 GB DDR4 modules should be here next year. I believe 32 GB unregistered, non-ECC DIMMs are eventually planned, though several years away. Reply
  • wireframed - Monday, August 04, 2014 - link

    Actually, there has already been 128GB modules (yes, modules, not kits) demoed. Since a lot of X99 boards will have 8 slots (2 slots per channel), you're looking at about a terabyte of memory on a board, provided the memory controller can handle it.

    Another thing people are forgetting: Even at the same speed, X99 is quad-channel, X79 was triple channel. So there's a 33% increase in bandwidth, even at the same speed. Add to that DDR4 starts off at a bit higher speed (looks like 2666MHz is going to be common right off the bat, and 3000MHz within reach for those who need it), and DDR4 boards ARE going to see an increase in performance for those applications that can use it.

    That said, I'm not sure most people are bandwidth constrained? I know most of what I do (3D rendering) taxes the CPU. Some things will benefit, but no where linearly with the bandwidth increase. I'm mostly looking at Haswell-E to get faster rendering, for those 10-20 hour renders. (Or 1 week+ for animations).

    Onboard video uses up a lot of bandwidth, but who buys a hex-core chip for that? (Especially considering they aren't going to have iGPU, AFAIK).
  • mapesdhs - Monday, August 04, 2014 - link

    Alas just because large capacity DIMMs exist, that doesn't mean a mbd can use them.
    Depends on the CPU, chipset, BIOS, etc. Indeed, Intel may greatly limit the max RAM
    of consumer HW-E in order to prevent their use from eating into XEON sales.

    Btw, X79 is quad-channel. Dunno where you got triple from, that was X58.

    Also, although the RAM clocks are up, the latencies are higher; I'm yet to be convinced
    DDR4/3000 @ CL15 (or whatever it is) will offer anything over DDR3/2400 @ CL10. It'll
    likely depend on the task, but the initial cost of DDR4 will also be horribly prohibitive.
    RAM has become so expensive in the last 18 months, the RAM cost alone may end up
    holding some people back from upgrading, even if the CPU & mbd costs are reasonable,
    eg. an 8GB 1600 kit costs 50% more now than what I paid for an 8GB 2133 kit in Feb/13.

    AE uses a lot of bandwidth; for those who can't afford costly XEON setups, X79 with an
    oc'd 6-core has been an ideal medium-cost intermediate build. I've constructed three such
    systems for people so far, all with combinations of Quadro 4K or K5K, multiple GTX 580s for
    CUDA, two with a 3930K, one with a 3960X, all running at 4.7GHz, two with 64GB RAM,
    one with 32GB RAM.

    Mine is a 3930K @ 4.7, 64GB/2133 (haven't yet optimised the RAM), 4x GTX 580 3GB
    (2X faster than a Titan Black for CUDA). It's fine for HD (though I've seen some scenes
    grab 40GB during processing), but to move up to 4K, that needs at least a doubling of
    the max RAM, preferably 4X.

    I know what you mean about rendering; friend of mine created a hefty AE/CUDA test
    which takes 15 mins to compute on my quad-580, the output looks like this (change the
    suffix to tga for the original 6MB Targa):

    As for CPU-heavy renders, I'm still looking for a suitable AE-based stress test. I found
    one benchmark, but it finishes in less than a minute, far too quick for useful testing.


    PS. The other big plus of X99 is the long awaited update to have lots more native Intel
    SATA3 ports, and other newer storage tech. The 3rd-party controllers used to provide
    non-Intel SATA3 on X79 boards are generally pretty awful.
  • Kevin G - Monday, August 04, 2014 - link

    The 128 GB modules are load reduced ECC modules which may not work in consumer systems. Intel has a habit of removing registered (and thus load reduced) support in their consumer systems. Reply

Log in

Don't have an account? Sign up now