Slashdot Mirror


Next-Generation DDR5 RAM Will Double the Speed of DDR4 In 2018 (arstechnica.com)

An anonymous reader quotes a report from Ars Technica: You may have just upgraded your computer to use DDR4 recently or you may still be using DDR3, but in either case, nothing stays new forever. JEDEC, the organization in charge of defining new standards for computer memory, says that it will be demoing the next-generation DDR5 standard in June of this year and finalizing the standard sometime in 2018. DDR5 promises double the memory bandwidth and density of DDR4, and JEDEC says it will also be more power-efficient, though the organization didn't release any specific numbers or targets. Like DDR4 back when it was announced, it will still be several years before any of us have DDR5 RAM in our systems. That's partly because the memory controllers in processors and SoCs need to be updated to support DDR5, and these chips normally take two or three years to design from start to finish. DDR4 RAM was finalized in 2012, but it didn't begin to go mainstream until 2015 when consumer processors from Intel and others added support for it. DDR5 has no relation to GDDR5, a separate decade-old memory standard used for graphics cards and game consoles.

39 of 77 comments (clear)

  1. What a waste by CODiNE · · Score: 1

    You passed up a First Post for THIS??

    --
    Cwm, fjord-bank glyphs vext quiz
  2. Insensitive clod! I'm still using DDR2 by Anonymous Coward · · Score: 1

    Insensitive clod! I'm still using DDR2

  3. Double Data Rate by subk · · Score: 1, Interesting

    Is the acronym "DDR" really still applicable? I remember back when it came out thinking we were going to see QDR and ODR come out next and next. I guess the naming convention folks just got lazy and started tacking on revision numbers.

    --
    Now, if you'll excuse me, I have backups to corrupt.
    1. Re:Double Data Rate by Anonymous Coward · · Score: 2, Informative

      DDR means that it transfers data on two clock edges. Clocks only have two edges (up and down).

      The number refers to other things about the protocol, such as voltage levels and allowable clock speeds.

    2. Re:Double Data Rate by SirSlud · · Score: 1

      "DDR5 promises double the memory bandwidth"

      What a stupid thing to whine about when it actually does double the data rate of the previous version.

      --
      "Old man yells at systemd"
    3. Re:Double Data Rate by TeknoHog · · Score: 3, Funny

      The number refers to other things about the protocol, such as voltage levels and allowable clock speeds.

      So it's 5 volts at 5 Hertz. Got it.

      --
      Escher was the first MC and Giger invented the HR department.
    4. Re:Double Data Rate by JustNiz · · Score: 1

      This is DDR Five. So I guess it should be FDR.

    5. Re:Double Data Rate by unixisc · · Score: 1

      Is the acronym "DDR" really still applicable? I remember back when it came out thinking we were going to see QDR and ODR come out next and next. I guess the naming convention folks just got lazy and started tacking on revision numbers.

      Very good point. When we moved from SDR to DDR, the whole idea was using both rising and falling edges of a clock as triggers for different parts of the memory circuit to reduce read or write times

      QDR or ODR might have made sense, but failing that, just call it SDR-2 for DDR, SDR-4, SDR-8, SDR-16 and so on

      On a different note, what's w/ the new orange titlebar on /.? The teal one was just fine.

    6. Re:Double Data Rate by Agripa · · Score: 1

      Is the acronym "DDR" really still applicable? I remember back when it came out thinking we were going to see QDR and ODR come out next and next. I guess the naming convention folks just got lazy and started tacking on revision numbers.

      DDR means that it transfers data on two clock edges. Clocks only have two edges (up and down).

      The number refers to other things about the protocol, such as voltage levels and allowable clock speeds.

      Plus there is a QDR (quad data rate) type of RAM which reads and writes on every clock edge so 4 bits per line are transferred on every clock cycle.

  4. Latency? by TeknoHog · · Score: 2

    So far, every new generation of Deutsche Demokratische Republik has doubled the capacity in bytes/s, while latency has been more or less constant. Will this be actually faster in that respect? Because having two 747s full of DVDs isn't any "faster" than a single one.

    --
    Escher was the first MC and Giger invented the HR department.
    1. Re:Latency? by FlyHelicopters · · Score: 1

      They can't change the latency, it hasn't changed in 15 years...

      The speed of electrons isn't going to get any faster, the RAM is too far from the CPU to be faster than it is today in terms of latency...

    2. Re:Latency? by LeftCoastThinker · · Score: 3, Interesting

      This is why I feel no great interest in upgrading to DDR4 (or DDR5). Once you have 64GB of RAM and are no longer touching the HDD for virtual memory, the speed changes between DDR3 and DDR5 for all but the most arcane applications is negligible, and video card is going to give you bigger bang for your buck in gaming/VR/AR/CAD/Video editing and the like.

      (Geek note: yes, yes I know it is not universal, thus the generalization. You spend $600 on DDR5 Ram and I will spend $350 on a video card and we will see who comes out ahead.)

      --
      If you disagree, please post your argument. (-1, Overrated) isn't your personal censorship tool for views you don't like
    3. Re:Latency? by willy_me · · Score: 1

      The future will eventually involve embedding RAM into the CPU core as ATI has done with their GPUs. Lower latency and can run at a lower voltage thereby requiring less power. They can also better control the shielding to allow for reduced voltage swing when communicating. Amazing how much power is required just to drive the signal lines between memory and the memory controller.

    4. Re:Latency? by KingMotley · · Score: 1

      Depends on how many library congresses worth of DVDs you need to move to China.

    5. Re: Latency? by Anonymous Coward · · Score: 1

      Cache?

  5. So, DDR5 will make DDR4...Faster?? by jfdavis668 · · Score: 2

    "Next-Generation DDR5 RAM Will Double the Speed of DDR4 In 2018" I think they need to learn to write better. I don't think the headline means what they think it means.

  6. Nothing revealed, not news by erice · · Score: 4, Interesting

    Article summary: "DDR5 is coming and it is going to be faster than DDR4"

    That's it. No technical details at all. Will it be point to point? If not, how many ranks? What voltage? What sort of termination? Will there even be DIMMs?
    You won't find answers to any of those questions in TFA or any page linked from the TFA. The only significant piece of information is confirmation that JEDEC has not given up on DDR.

    1. Re:Nothing revealed, not news by Anonymous Coward · · Score: 1

      Article summary: "DDR5 is coming and it is going to be faster than DDR4"

      That's it. No technical details at all. Will it be point to point? If not, how many ranks? What voltage? What sort of termination? Will there even be DIMMs?
      You won't find answers to any of those questions in TFA or any page linked from the TFA. The only significant piece of information is confirmation that JEDEC has not given up on DDR.

      Still, if it is true it is a good thing. Fast code needs to take advantage of the cpu cache, not be continually replacing it. A glance seems to show that L1 cache is around 100 times faster than DDR4 and around 50 times faster than L2 and maybe 25 as L3. Now if you install say 4 ddr5 modules, than might be 25, 13, and 7 times faster. I'm slightly suspicious about those numbers, but assuming that is right, well doubling the rate reduces the need for L3 cache. At some point it would become better to toss the L3 in favor of more L2 or L1, or just toss it in favor of a cheaper computer or adding something elsewhere. This doesn't appear to be that point.

      I'm guessing the greater value would be integrated video. Doubling the bandwidth to ram might have a huge impact for integrated video.

      Still it would be interesting if you knew, from a coding point of view, that it doesn't matter how stuff is stored in memory, you can get it all in the same amount of time. That will probably never be entirely true since your bound to load things by some kind of block size, but it may become at least partially true.

    2. Re:Nothing revealed, not news by phantomfive · · Score: 1

      The standard still isn't released, and won't be released until 2018. The actual RAM might not make it here until 2020.

      The main news (afaict) is that DDR5 will have double the bandwidth of DDR4. It's also going to have lower power requirements yet, and since the standard hasn't been finalized yet, grant you a pony.

      --
      "First they came for the slanderers and i said nothing."
    3. Re:Nothing revealed, not news by Dogtanian · · Score: 2

      Article summary: "DDR5 is coming and it is going to be faster than DDR4"

      That's it. No technical details at all. Will it be point to point? If not, how many ranks?

      Well, it's one better, isn't it? It's not four. You see, most RAMs, you know, will be sticking at four. You're on four here, all the slots. Where can you go from there? Where?

      Nowhere. What they do is, if they need that extra push over the cliff, you know what they do?

      Put it up to five. One better.

      --
      "Slashdot - News and Chat Sites Deviant". (Click "homepage" link above for details).
    4. Re:Nothing revealed, not news by ChrisMaple · · Score: 1

      The bigger memory is, the slower it is. You can't just arbitrarily double the size of L1 or L2 without making it slower. CPU designers go to a lot of trouble to optimize the size, speed, and number of levels of cache for typical workloads.

      --
      Contribute to civilization: ari.aynrand.org/donate
  7. Re: Insensitive clod! I'm still using DDR2 by HumanWiki · · Score: 1

    My FPM would like to talk to you both.

  8. Might double, but CAS latencies jump with it by Khyber · · Score: 2

    Thus the actual increase in performance is very, very, very little.

    --
    Still waiting on Serviscope_minor to wake up to fucking reality and realize that Jessica Price isn't going to fuck him.
    1. Re:Might double, but CAS latencies jump with it by nadaou · · Score: 2

      > Thus the actual increase in performance is very, very, very little.

      Nope. CAS latencies are measured in clock cycles, so what you are seeing is the latency for the first read remaining at about 9ns. As the frequency increases the number of cycles it takes before 9ns has passed increases as well. Overall observable latency stays about the same for the first read, but you gain from the higher frequency on subsequent reads.

      --
      ~.~
      I'm a peripheral visionary.
    2. Re:Might double, but CAS latencies jump with it by Khyber · · Score: 1

      You don't gain when you've got the RAM slots way away from the processor, though, like in quite a few of the larger ATX motherboards.

      Really, the only true gain to be seen is having the RAM built onto the die itself.

      --
      Still waiting on Serviscope_minor to wake up to fucking reality and realize that Jessica Price isn't going to fuck him.
  9. Thats a neat trick. by JustNiz · · Score: 4, Funny

    >> DDR5 RAM will double the speed of DDR4 in 2018

    Thats cool! Will my DDR4 go back to normal speed in 2019?

    1. Re:Thats a neat trick. by JustNiz · · Score: 1

      Well, being English I think I can reasonably assume I'm a native English speaker. And no, correct grammar is not "inappropreate" (sic) for this site.

  10. Re: Insensitive clod! I'm still using DDR2 by JustNiz · · Score: 1

    What is this new-fangled Dynamic memory you speak of? I'm still using static me#@$@$^*

  11. You don't need to wait! by JustNiz · · Score: 4, Funny

    > DDR5 RAM will double the speed of DDR4 in 2018

    I heard you can set your system clock ahead to 2018 and your RAM will go double-speed today.

  12. Oh by SuperKendall · · Score: 4, Funny

    I heard you can set your system clock ahead to 2018 and your RAM will go double-speed today.

    So THAT"S what overclocking means!

    --
    "There is more worth loving than we have strength to love." - Brian Jay Stanley
  13. That sucks by frovingslosh · · Score: 1

    I was hoping that they could half the speed of the next generation RAM. Instead they are doubling it?

    --
    I'm an American. I love this country and the freedoms that we used to have.
  14. More importantly than speed... by Gravis+Zero · · Score: 1

    will DDR5 still be susceptible to Rowhammer attacks? Also, will the increased speed make it even more susceptible? They barely escaped a cataclysm last time, so one would hope they addressed the root cause.

    --
    Anons need not reply. Questions end with a question mark.
    1. Re:More importantly than speed... by slew · · Score: 1

      will DDR5 still be susceptible to Rowhammer attacks? Also, will the increased speed make it even more susceptible? They barely escaped a cataclysm last time, so one would hope they addressed the root cause.

      The root cause of Row Hammer is physics: reading a row causes adjacent row capacitors to degrade at a faster rate than typical.

      You can refresh the memory even more conservatively to eliminate these statistical outliers (reducing average performance and increasing power) or come up with a more nuanced approach (e.g., doing Targeted Row Refresh or TRR). In TRR mode the DRAM chip tracks row activations to identify a row that is being targetted and prepends the two adjacent rows to the Refresh sequence so they can be refreshed before the normal Refresh sequence gets around to them so they don't degrade too much.

      Given that many vendors have added some non-standard support for TRR mode on DDR4 (this feature was added as an optional feature for LPDDR4 and some vendors added it to their DDR4 chips even though it wasn't part of the DDR4 spec), I suspect that is the direction they are headed will be TRR on DDR5.

    2. Re:More importantly than speed... by Gravis+Zero · · Score: 1

      The root cause of Row Hammer is physics: reading a row causes adjacent row capacitors to degrade at a faster rate than typical.

      Close but not quite. The root cause is reading a row causes adjacent row capacitors to degrade at a faster rate than they can be replenished. It is a result of physics and a lack of foresight.

      One solution is to add a tiny amount of cache that will prevent higher than typical access rates from draining the adjacent row's capacitors but there are other solutions that would work just as well.

      --
      Anons need not reply. Questions end with a question mark.
    3. Re:More importantly than speed... by ChrisMaple · · Score: 1

      Isn't cache-on-RAM part of what made RAMBUS such a disaster?

      --
      Contribute to civilization: ari.aynrand.org/donate
  15. Sweet, even more HBM! by Entrope · · Score: 1

    That's Heaping Bullshit Marchitecture, for those not in the know.

  16. Re: Insensitive clod! I'm still using DDR2 by slew · · Score: 1

    What is this new-fangled Dynamic memory you speak of? I'm still using static me#@$@$^*

    It's like Magnetic-Core memory, except it forgets... Segmentation fault (core dumped)

  17. Simply amazing by mi · · Score: 1

    Next-Generation DDR5 RAM Will Double the Speed of DDR4 In 2018

    This is awesome technology... I mean, to double the speed of all the DDR4 memory already sold and installed in computers world-wide — without even touching them. Pure magic...

    --
    In Soviet Washington the swamp drains you.
  18. Re: Insensitive clod! I'm still using DDR2 by HumanWiki · · Score: 1

    Punch card. Abacus. Stones. :)