back to article Double-DIMMed XPoint wastes sockets

A Xitore white paper compares coming XPoint DIMMs and Xitore's own flash DIMMs, and claims each XPoint DIMM needs a companion DRAM cache DIMM, obviously halving XPoint DIMM density. The startup has its own tech to push – NVDIMM-X – but, even so, is revealing about XPoint DIMMery. Doug Finke, director of product marketing at …

  1. Steve Davies 3 Silver badge

    And for us mere mortals...

    seeing this on any kit we are likely to get our hands on is about as certain as Hens sprouting Canine teeth.

    TBH, we have seen so many 'magic solution' storage vendors come and go over the years here on El Reg that I for one am getting a bit wary of these companies and their claims/promises.

    IMHO is that all they want is to get bought out by a bigger org so that the directors can make a killing and then move on to their next 'big thing'.

  2. Joerg

    Then what ?

    1TB XPoint DIMM and a 16GB DIMM RAM cache would be more than enough performance wise.

  3. Anonymous Coward
    Anonymous Coward

    Whats up with those numbers?

    "((950,000 x 1) = 950,000) + ((500 x 5) = 2,500) = 952,500 time units

    The average time per access is 0.9525 time units.

    Let's employ flash DIMMs instead of XPoint ones, with an access time of 50 time units, 10 times slower, and use the same DRAM caching scheme and hit rate. What is the total access time for 1 million IOs?

    ((950,000 x 1) = 950,000) + ((500 x 50) = 25,000) = 975,000 time units

    The average time per access is 0.975 time units. "

    I thought it was supposed to be 1 MEEEELLION IOs?

    Why is it 950000 in RAM and only 500 in NVRAM?

    Would have thought that the give away it was wrong would have been the average access time unit when combining RAM and NVRAM was faster than the fastest either could do.

    Or have I misinterpreted the numbers?

    1. Pascal

      Re: Whats up with those numbers?

      Yeah I think there's a math glitch here?

      The author is counting 500 cache misses instead of 50,000.

      It should be:

      ((950,000 x 1) = 950,000) + ((50,000 x 5) = 250,000) = 1,200,000 time units

      vs

      ((950,000 x 1) = 950,000) + ((50,000 x 50) = 2,500,000) = 3,450,000 time units

      1. Anonymous Coward
        Anonymous Coward

        Re: Whats up with those numbers?

        Which then changes the conclusion dramatically, as it was supposed to be only 2.36% difference in speed, when in actuality it was almost 200%.

      2. Jim O'Reilly

        Re: Whats up with those numbers? A bigly problem!!!

        If I were designing this, I would have two flash DIMMs instead of one X-Point DIMM plus its DRAM DIMM. With parallel accesses, this effectively halves the time for 1 million IOPS. X-Point isn't fast enough to be worth 5x!

      3. dougfinke

        Re: Whats up with those numbers?

        The rough performance calculation that was shown did miss one other key point. The NVDIMM-X implements a command queueing capability similar to the NCQ capability introduced by Seagate for SATA HDDs in 2004. This allows for execution out-of-order and can reduce the impact of the slower accesses at the system level. DRAM DIMMs have never had this capability and it is not known whether the 3D Xpoint DIMM has this or not. If it doesn’t, then the performance advantage for 3D Xpoint will be a lot less.

        Also, the Intel spokesperson may not have been aware, but the two socket 3D Xpoint NVDIMM architecture was shown at the August Intel Developer Forum session in the paper titled: “Wicked Fast Storage and Beyond”.

  4. Anonymous Coward
    Anonymous Coward

    Good image selection...

    .... always thought UK sockets are just a waste of space...

    1. Anonymous Coward
      Anonymous Coward

      Re: Good image selection...

      Out of curiosity, what would you prefer?

    2. Anonymous Coward
      Anonymous Coward

      Re: Good image selection...

      safer than the Euro ones ;)

      1. Anonymous Coward
        Anonymous Coward

        Re: Good image selection...

        "safer than the Euro ones"

        The Europlug is only really potentially dangerous because of the design of the BS1363 socket, and trailing adaptors, which means that one pin of the two pin version can be pushed into the live BS1363 hole leaving an exposed live pin. Pity, because but for that we could have 12 way adaptors in the space currently taken up by a 4-way BS1363, plus plugs that are step on safe.

        Off topic but still worth mentioning; it's another thing that was actually a good idea but sabotaged (unintentionally may I add) by the UK. Equipment manufacturers really liked it.

        1. Snar

          Re: Good image selection...

          To be able to force a Europlug into a BS1363 socket, you need to use something to lower the shutters covering the Live and Neutral by forcing something into the Earth terminal. If you try and push a Europlug into a BS1363 socket without defeating the Earth, then it won't engage.

          So if you have the basic intelligence to understand how to defeat the shuttering and lack the common sense to plug the appliance in correctly and use just one pin in the defeated socket, then:

          a) The appliance won't work.

          b) You have a problem with your sight because one of the pins of the Europlug is not in the socket.

          c) A Darwin Award is your personal goal.

          Or have I missed something here?

  5. Anonymous Coward
    Anonymous Coward

    I see the they have tried changing the article, still missing a 0 off of both, 5000 instead of 50000

    Just copy what Pascal wrote :p

  6. Anonymous Coward
    Anonymous Coward

    Write or read?

    I believe the author has it sideways - write cache is about returning in 1 time unit always, while the write takes (whatever, 5 time units). The author treats it as if it were read caching and others have noticed the math error, but the real error is, that's not what the quote is about. (it may yet be an issue but it's not raised by the quoted text)

  7. Anonymous Coward
    Anonymous Coward

    I'd like to see a slice of memory ...

    RAM-ed in a UK socket.

    but that would be just plain silly :)

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon