Become a fan of Slashdot on Facebook

 



Forgot your password?
typodupeerror
×
Hardware

Micron Displays Next-Gen LPCAMM2 Modules For Laptops At CES 2024 28

At CES 2024 this week, Micron demonstrated its next-gen LPCAMM2 memory modules based on LPDDR5X memory. Not only are they smaller and more powerful than traditional SODIMMs, they can be "serviced during the manufacturing process and upgraded by the user," says Micron. Tom's Hardware reports: Micron's LPCAMM2 are industry-standard memory modules that will be available in 16 GB, 32 GB, and 64 GB capacities as well as with speed bins of up to a 9600 MT/s data transfer rate. These modules are designed to replace conventional SODIMMs as well as soldered-down LPDDR5X memory subsystem while offering the best of both worlds: flexibility, repairability, and upgradeability of modular memory solutions as well as high performance and low power consumption of mobile DRAM. Indeed, a Micron LPCAMM2 module is smaller than a traditional SODIMM despite the fact that it has a 128-bit memory interface and up to 64 GB of LPDDR5X memory onboard. Needless to say, the module is massively smaller than two SODIMM memory sticks that offer a 128-bit memory interface both in terms of height and in terms of physical footprint.
This discussion has been archived. No new comments can be posted.

Micron Displays Next-Gen LPCAMM2 Modules For Laptops At CES 2024

Comments Filter:
  • Nice try, engineers! But I am sure Apple will find a way to subvert the intentions of the designers and turn it into a non-serviceable, non-upgradable, and far overpriced part.

  • Every discussion I've seen of this jumps on Apple. The ARM based Mx processors use on package memory (HBM). You have to replace the whole SoC/processor to upgrade the memory. This actually results in much faster memory, so they're not just doing it for lock in (though I'm sure they're not complaining about that).
    • by Lehk228 ( 705449 )
      end result is that as long as you are within it's memory capabilities it's insanely efficient but i do wish they offered an additional memory slot for slower system ram
      • by tepples ( 727027 )

        i do wish they offered an additional memory slot for slower system ram

        Agreed. Even if slower slot RAM is used only for a swap file, it can still fill a role in the memory hierarchy.

      • Itâ(TM)s called SSD, which is over a sufficiently fast interconnect, it effectively is as fast as DDR4 RAM. The problem with DDR5 is that itâ(TM)s so fast, not a single manufacturer has been able to put it in SO-DIMM format, because the distance from the left to the right of the module is too far for the speed of light. So you either put it on board or have to invent a non-standard like this CAMM format which is still inferior to the DDR6 HBM on GPU and modern ARM processors.

        • It's called SSD, which is over a sufficiently fast interconnect, it effectively is as fast as DDR4 RAM.

          The difference is that a DDR4 RAM disk doesn't wear out as fast as multi-level-cell NAND flash memory when you put a swap file on it.

          • by guruevi ( 827432 )

            Wearing out in high quality SSD hasn't been a problem for over a decade, even if you used it as RAM.

            • This surprised me. I asked elsewhere and was told that anything more dense than SLC (1 bit per cell) has noticeablly lower write durability if used as swap for several years. And because SSD makers have long since replaced SLC with MLC or even more dense, I doubt any PC maker bundles a "high quality SSD" anymore. I'm open to evidence to the contrary, evidence that an MLC SSD swaps as fast a DDR4 RAM disk and has negligible wear even if used for a swap workload over its service life.

              • by guruevi ( 827432 )

                Apple and other enterprise/datacenter vendors use high speed intake buffers on-die (or swap space in the case of Apple) by treating the area as SLC, then copy permanent data at a later time to the MLC. With wear leveling and other controller 'tricks', modern datacenter/enterprise SSDs can be fully written 3x per day and stay within spec (you can write 6TB to a 2TB SSD in a 24 hour period).

    • The ARM based Mx processors use on package memory (HBM).

      I thought that too, but it's not HBM, it's a bunch of package on package LPDDR in parallel.

  • Has anyone opened up a notebook recently? 50% empty space.

    • Erm I have. And mine is 0% empty space, given that all "empty" space is taken up by a battery. I'm not sure what on earth you are buying if you have 50% empty space. That sounds like a pointlessly large device.

      But in any case there's more to this than just space. The Dell CAMM patent shows quite clearly why this is the future direction for memory (for those people who don't want chips soldered directly on the board or on the CPU die). SODIMMs are getting to the point where signalling and timing issues are a

    • Firstly yes I have and no it isn't. Secondly SODIMM has reached the end of the line electrically. It cannot be made to work with LPDDR memory. Personally I want low power notebooks to be upgradeable too since RAM is often the first thing to age out.

  • By pioneering the CAMM design and rushing it to market, their standard was made obsolete literally within 1.5 years with JEDEC releasing CAMM2 in December. On the up side it's good a few DELL customers ironed out the bugs in the design.

    Hopefully this puts an end to soldered on memory nonsense.

    • by Khyber ( 864651 )

      "Hopefully this puts an end to soldered on memory nonsense."

      Fuck no. Intel just showed off new processors with the RAM on-package. we're getting closer to removal of the RAM slot, entirely.

      • There's a difference between RAM on die and RAM on motherboard. There will always be a requirement for component level compartmentalisation, it really doesn't matter what "Intel showed off". They don't need to show off anything. RAM on die or RAM on SoC is well over a decade old. The question is whether it is accepted outside of the Apple / integrated device crowd.

Every nonzero finite dimensional inner product space has an orthonormal basis. It makes sense, when you don't think about it.

Working...