Slashdot Mirror

← Back to Stories (view on slashdot.org)

Intel Launches Optane DIMMs Up To 512GB (anandtech.com)

Posted by msmash on from the it's-here dept.

Intel announced the availability of its long-awaited Optane DIMMs Wednesday, bringing 3D XPoint memory onto the DDR4 memory bus. From a report: The modules that have been known under the Apache Pass codename will be branded as Optane DC Persistent Memory, to contrast with Optane DC SSDs, and not to be confused with the consumer-oriented Optane Memory caching SSDs. The new Optane DC Persistent Memory modules will be initially available in three capacities: 128GB, 256GB and 512GB per module. This implies that they are probably still based on the same 128Gb 3D XPoint memory dies used in all other Optane products so far. The modules are pin-compatible with standard DDR4 DIMMs and will be supported by the next generation of Intel's Xeon server platforms.

31 of 54 comments (clear)

  1. I'm not up on all the jargon by H3lldr0p · · Score: 3, Interesting

    So is this one step closer to non-volatile RAM or not?

    I mean, are we finally going to bridge that gap between storage and RAM so that everything is finally moving at bus speed or is that still the fevered dreams of ultra-systems-on-a-chip?

    1. Re:I'm not up on all the jargon by swb · · Score: 2

      If I remember the hype correctly, Optane is supposed to be some multiple faster than SSD with superior durability. My guess this is some kind of gimmick to break through disk bus bandwidth limits but without using NVMe slots.

      A lot of servers seem to have a bunch of empty DIMM slots, so it maybe makes sense to sell Optane that can be plugged into existing DIMM slots.

      What's not clear is how they plan to present this to the computer -- is it going to be as a disk device, or is it just going to be mapped to RAM? My understanding is that Optane is still much slower than RAM, so it would seem to make sense to present it as a disk device somehow. I'm not sure any OS would even have a way of dealing with "these memory pages act like non-volatile storage, not like RAM".

      I'm also curious what having a couple of TB of Optane as storage on your memory bus would do for memory access times. I think we're already cursed going to main memory as it is, having that bus tied up with large Optane traffic sounds like it would make memory access worse.

      Maybe Intel's plan is to just use DDR as some kind of backplane/interconnect and plans multiple memory busses in some future chipset release that allows for Optane and DRAM to not get in each other's way.

    2. Re:I'm not up on all the jargon by GerryGilmore · · Score: 1

      It's already NVRAM. Yep, just the old core memory boards, but without the need to rewrite after every read.

    3. Re:I'm not up on all the jargon by lexman098 · · Score: 1

      are we finally going to bridge that gap between storage and RAM so that everything is finally moving at bus speed or is that still the fevered dreams of ultra-systems-on-a-chip?

      Why would you want that? There's multiple levels of busing for a reason. If you can access main storage at the same speed as RAM then either:
      A) your main storage is too expensive or
      B) your RAM is too slow
      Either of these situations is ripe for optimization.

    4. Re:I'm not up on all the jargon by swb · · Score: 1

      It would be interesting to have several terabytes of non-volatile memory that was accessible at DRAM speed. You would have a single unified memory space for data and executables. No waiting to move data from storage to RAM.

      You could turn off the power and turn it back on and regain the exact state of execution without any of the delays associated with hibernation or the power consumption of sleep.

    5. Re:I'm not up on all the jargon by Tailhook · · Score: 1

      Not.

      It's nonvolatile storage with a DIMM connector. Intel has always positioned "3D XPoint" as either a nonvolatile cache for slower storage or an independent storage medium; they aren't claiming this replaces RAM. Latency and bandwidth are far behind contemporary DDR regardless of what edge connector you put on it, and no one is claiming otherwise.

      Maybe in the fevered dreams of some tech writer a future refinement of 3D XPoint can somehow compete with capacitors, but don't hold your breath waiting for it.

      --
      Maw! Fire up the karma burner!
    6. Re:I'm not up on all the jargon by tlhIngan · · Score: 1

      What's not clear is how they plan to present this to the computer -- is it going to be as a disk device, or is it just going to be mapped to RAM? My understanding is that Optane is still much slower than RAM, so it would seem to make sense to present it as a disk device somehow. I'm not sure any OS would even have a way of dealing with "these memory pages act like non-volatile storage, not like RAM".

      Every OS has this ability. We call it a "RAM Disk". Even better, a lot of systems now can have persistent RAM disks as long as power is running

      Sure some modifications will be needed - like the kernel recognizing that it shouldn't map it in as normal memory (but it can, but it might destroy RAM disk contents), but to be aware of it. The OS RAM disk driver can then be told those physical DIMMs are persistent storage RAM disks and there you go.

      The BIOS/UEFI will probably mark it as special memory when it passes control to the OS loader

    7. Re:I'm not up on all the jargon by MightyYar · · Score: 1

      This is what the original Palm Pilots did. Eventually flash technology moved forward and they had to make awkward kludges to support it - but when it was RAM-only it made for a very responsive and easy to use device.

      --
      W..w..W - Willy Waterloo washes Warren Wiggins who is washing Waldo Woo.
    8. Re:I'm not up on all the jargon by lexman098 · · Score: 1

      No waiting to move data from storage to RAM.

      This is a long-long-long way off with even mainstream high-end components if it ever happens at all. My point above was that if you have enough space in your RAM to store everything you need then your RAM is too slow or expensive. There will always be a faster bus or a new DDRx standard that's bleeding edge and would cost too much to support the increasing file sizes needed for modern programs. A cascade of buses decreasing in speed and cost will always be more efficient. The good news is that the time you wait *now* to move something between RAM to the CPU cache (another buffer in the cascade) will *eventually* be equal to the time you wait to move something from storage to RAM.

    9. Re:I'm not up on all the jargon by TomGreenhaw · · Score: 1

      Yes, its a big step in that direction. I think these will be used on motherboards and OSs designed for them, but using DIMMS would provide vastly more bandwidth for non volatile storage. Using these instead of/in addition to SATA or M2 will mean instant on computing, massive in memory database servers at a reasonable cost and systems that are far more tolerant to power issues and unexpected shutdowns.

      --
      Greed is the root of all evil.
    10. Re:I'm not up on all the jargon by Bing+Tsher+E · · Score: 1

      That was part of the sheer beauty of a PalmOS device. You could sync it to desktop, then take any other PalmOS device you had, put batteries in it and 'sync' it to the desktop to pull the entire image back onto a separate device.

      Plus, on the better (earlier) versions of the Palm desktop, you could zip up the whole folder on your PC the Palm Desktop was installed in and lug it around to different computers, unzip it, and yes, run it to sync a Palm device to it to pull the PalmOS setup onto a device.

      Anybody who knew what they were doing and cared to keep it still has that zipfile stuck away somewhere. I could pull it out and put it on a 32 bit Windows box and have my calendar and todo list and everything back on my Palm 3 easily.

      It's a shame there isn't something that easy and complete available on Android. Oh yeah, there probably is. There are probably 485 'apps' that will do that, or 57-83% of that, if you futz enough with it.

    11. Re:I'm not up on all the jargon by omnichad · · Score: 1

      However in a server environment it really has no practical use unless it's being used as network storage.

      I dunno. If you could suspend an entire VM to optane storage (RAM copy) until a network request comes in, and you could switch quickly, there is some real power savings potential. Especially if you could execute directly from optane while paging back to RAM. This would only work for intermittent use VMs, but you could shut down entire cores on a server that would otherwise be wasted.

      You could argue that any server that could use this was under-provisioned anyway. I'm not so sure.

    12. Re:I'm not up on all the jargon by swb · · Score: 1

      AFAIK, Windows doesn't have any native RAM disk capability. There are third party applications that can add this, but conceptually it seems a lot different than having what amounts to disk storage in RAM slots.

      A system running a RAM disk doesn't really care what pages get used for RAM disk and the memory control of the kernel treats all RAM pages equally.

      I don't doubt Linux has some kind of special memory modes evolved from SoC usage or something where flash and RAM are already mapped to a common memory space. But it also seems likely that this is mostly a hardware gimmick to eliminate disk controllers and the applications need some intelligence to know whether they're writing to flash pages or RAM pages unless the kernel memory manager automatically makes a disk out of flash pages and only exposes real RAM to applications.

      It strikes me that without a special memory control manager that knows about flash pages vs. ram pages, a generic OS booted on an optane equipped system would just not work well unless it treated those pages differently.

    13. Re:I'm not up on all the jargon by torkus · · Score: 1

      Not.

      It's nonvolatile storage with a DIMM connector. Intel has always positioned "3D XPoint" as either a nonvolatile cache for slower storage or an independent storage medium; they aren't claiming this replaces RAM. Latency and bandwidth are far behind contemporary DDR regardless of what edge connector you put on it, and no one is claiming otherwise.

      Maybe in the fevered dreams of some tech writer a future refinement of 3D XPoint can somehow compete with capacitors, but don't hold your breath waiting for it.

      While they aren't (currently) trying to replace RAM, their Optane line is certainly not just a cache or faster alternative for flash storage. Not at all. They are looking at use cases where the quantity of memory is more important than the absolute speed. With the optane DIMMs they can populate more RAM in a single DIMM than most servers can support fully populated otherwise.

      If you aren't a use case for max-RAM servers than optane DIMMs have ver little utility for you. If you are, then this is an enormous potential and cost saver.

      --
      You can get rich if you own a politician, but you have to be rich to buy one in the first place.
  2. Re:who cares who cares by Anonymous Coward · · Score: 1

    I care, because unlike the politifeces posted far too often on this site, this is ACTUALLY TECH NEWS.

    Depending on how the year goes, I may be able to requisition a Xeon with this kind of RAM for work, so the actual success in converting the theory to physical RAM is worthy of note.

  3. and amd chips have loads of pci-e for ssd hot swap by Joe_Dragon · · Score: 1, Informative

    and amd chips have loads of pci-e for ssd;s that can be hot swapped.

  4. Naming conventions by PraiseBob · · Score: 3, Insightful

    Why is Intel naming all of their new memory products Optane? Near as I can tell, Optane might refer to their m2 chip extension of SRT to boost mechanical drives, or might be a couple different lines of super fast SSD, or might be a dimm now? Somebody needs to talk to their branding & marketing dept, to stop calling everything the same name. It really makes looking up the specs confusing.

    1. Re:Naming conventions by swb · · Score: 1

      You're not supposed to look at the specs, you just need to purchase at least 3 variations of Optane for every new server.

    2. Re:Naming conventions by Anonymous Coward · · Score: 2, Informative

      they are naming all of their 3d xpoint derived products optane...

    3. Re:Naming conventions by Anonymous Coward · · Score: 4, Funny

      Why is Intel naming all of their new memory products Optane?

      You'll be able to buy RAM with Optane 87, Optane 89, or Optane 91. For high performance PC's you can get Optane 98 at specialty shops.

    4. Re:Naming conventions by TeknoHog · · Score: 1

      Somebody needs to talk to their branding & marketing dept, to stop calling everything the same name. It really makes looking up the specs confusing.

      Whoa there, you can always buy a Core or a Pentium if you don't want Optane.

      --
      Escher was the first MC and Giger invented the HR department.
    5. Re:Naming conventions by Anonymous Coward · · Score: 1

      Why is Intel naming all of their new memory products Optane?

      It's from the Marketing department, it doesn't have to make sense.

    6. Re:Naming conventions by thegarbz · · Score: 3, Informative

      Why is Intel naming all of their new memory products Optane?

      This may blow your mind, but it's because all their new memory products are based on the Optane (3D Xpoint) technology. Their marketing department is doing just fine showing that Intel is putting proprietary memory technology in their products.

      I don't think anyone is at risk of having the SRT component confused with a memory DIMM just because it has Optane in the name. It's like complaining about the fact that new harddrives all have "solid state" in the title, they also cross a wide variety of form factors and performance metrics.

    7. Re:Naming conventions by ausekilis · · Score: 1

      I can't find the gas cap on my computer. How do I know which Optane to put in it? Will putting in a higher Optane make it go faster? If I put in a higher Optane do I have to change my mineral oil every 5000 hours?

  5. Re:IN-sufficient! by thegreatbob · · Score: 1

    They will always find a way to use it, just as a gas tends to expand to fit the volume of its container. A different kind of vaporware, I guess?

    --
    There is no XUL, only WebExtensions...
  6. Re:Naming conventions - Pentane is taken by mnemotronic · · Score: 2

    Pentane is used for lava targets. Heptane and Septane are already spoken for. Enneatane don't fire nobody's rocket, not nobody, not nohow. Hentriacontadictakitane is available.

    --
    The Russians have won. They have made the world a cesspool of distrust, greed, fear and hate.
  7. Re:who cares who cares by Tough+Love · · Score: 1

    Not you. And nobody cares about that.

    --
    When all you have is a hammer, every problem starts to look like a thumb.
  8. Re:who cares who cares by Tough+Love · · Score: 1

    this kind of RAM

    NVRAM.

    --
    When all you have is a hammer, every problem starts to look like a thumb.
  9. write utilization tells the story by epine · · Score: 1

    The 60 DWPD warranty over 5 years pretty much sums this up.

    A single DDR4 interface offers about 20 GB/s bandwidth. Over five years, that's 3.15 exabytes, under continuous write traffic (LIGO could generate this access pattern, I suspect).

    1 TB * 60 writes/day * 5 years = 109 petabytes.

    (The article only implies 5 years rather than 3 years, but I chose the generous figure.)

    I work that out at around a 3.5% write bandwidth utilization over the warranty lifecycle (less than 2% the 512 GB part, less than 1% for the 256 GB part).

    Clearly this is not your father's DRAM replacement.

    But I can sure imagine this being the sweet spot for a ZFS NAS server's L2ARC (if cache occupancy is fairly stable).

    And there's probably some substantial latency optimizations to be had for not going over NVMe (an interface which is pretty busy writing your ZIL to something with a lot more write depth, anyway).

    1. Re:write utilization tells the story by epine · · Score: 1

      Crap, I got all the capacities wrong by a factor of two, not that it changes much of anything.

  10. Re:who cares who cares by AHuxley · · Score: 1

    When an OS, the application and user find a task that needs lots of not RAM speed memory then it works.
    Once such a task is discovered the OS, app and user have to keep a task within the limits of what Optane can do.
    Something software aware thats way too big for RAM, needs more speed than a new SSD.
    Video work? A huge computer game loading? Graphics and an art application?
    Some task has to fit right? Between really fast RAM and the huge fast SSD.

    --
    Domestic spying is now "Benign Information Gathering"