Slashdot Mirror

← Back to Stories (view on slashdot.org)

Samsung Unveils V-NAND High Performance SSDs, Fast NVMe Card At 5.5GB Per Second

Posted by samzenpus on from the greased-lightning dept.

MojoKid writes: Sometimes it's the enterprise sector that gets dibs on the coolest technology, and so it goes with a trio of TCO-optimized, high-performance solid state drives from Samsung that were just announced, all three of which are based on three-dimensional (3D) Vertical NAND (V-NAND) flash memory technology. The fastest of bunch can read data at up to 5,500 megabytes per second. That's the rated sequential read speed of Samsung's PM1725, a half-height, half-length (HHHL) PCIe card-type NVMe SSD. Other rated specs include a random read speed of up to 1,000,000 IOPS, random write performance of up to 120,000 IOPS, and sequential writes topping out at 1,800MB/s. The PM1725 comes in just two beastly storage capacities, 3.2TB and 6.4TB, the latter of which is rated to handle five drive writes per day (32TB) for five years. Samsung also introduced two other 3D V-NAND products, the PM1633 and PM953. The PM1633 is a 2.5-inch 12Gb/s SAS SSD that will be offered in 480GB, 960GB, 1.92TB, and 3.84TB capacities. As for the PM953, it's an update to the SM951 and is available in M.2 and 2.5-inch form factors at capacities up to 1.92TB.

61 comments

  1. How much? by sectokia · · Score: 2

    Even I have my limits.

    1. Re:How much? by beelsebob · · Score: 1

      Roughly $2.50 per gig.

    2. Re:How much? by Anonymous Coward · · Score: 0

      So $ 9,500 for the 3.84TB one? That's insane.

    3. Re:How much? by beelsebob · · Score: 2

      It's very expensive. On the other hand, a nearly 4TB SSD that reads at 5.5GB/s is a somewhat exceptional SSD.

    4. Re:How much? by Anonymous Coward · · Score: 1

      Even I have my limits.

      And when speaking of anything in the latest-and-greatest category, since when has the price tag been anything south of what-the-fuck-man?!?

      This is also why basketball shoes are now sold like cars, with the 2015 model commanding the most premium (or perhaps the 2016 models are now hitting the floor)

    5. Re:How much? by Anonymous Coward · · Score: 0

      Imagine a beowulf cluster of 4TB SSD

    6. Re:How much? by knightghost · · Score: 2

      $10k is extremely cheap when you are looking at speeding up a multi-TB database with a 100+ queries per second 24/7.

    7. Re: How much? by bill_mcgonigle · · Score: 1

      it's what I paid for a 1.2TB raid array 10 years ago, based on PATA hard disks.

      This one is three times as big and four orders of magnitude faster. That's an amazing evolution.

      --
      My God, it's Full of Source!
      OUTSIDE_IP=$(dig +short my.ip @outsideip.net)
    8. Re:How much? by blackiner · · Score: 1

      Hopefully the mere fact that this thing even exists means that lower capacities will start to fall in price. They are obviously getting better at making SSDs... a few years back, the mere idea of a 4TB SSD would be unthinkable.

    9. Re:How much? by Anonymous Coward · · Score: 5, Funny

      Imagine a beowulf cluster of 4TB SSD

      Imagine a beowulf cluster of ancient redundant slashdot memes, written on scrap lumber and jammed up your ass.

    10. Re:How much? by ShanghaiBill · · Score: 4, Informative

      $10k is extremely cheap when you are looking at speeding up a multi-TB database with a 100+ queries per second 24/7.

      The only alternative without a performance hit, would be battery backed RAM. 4TB of battery backed RAM would cost about four times as much.

    11. Re: How much? by Anonymous Coward · · Score: 0

      Are you sure about 10 years ago (2005) and about the price? I found my old emails from 2005 where I was putting together a barebones PC with PATA drives, Athlon64 CPU, and about 1GB RAM. The base machine (sans disk) was about $500 and the disks were under $0.50/GB at sizes like 160GB or 320GB so it seems like a 1.2TB RAID5 server should have cost at most $2000 but could have been as low as $1200 if budgeted carefully.

    12. Re:How much? by blang · · Score: 5, Insightful

      That's dirt cheap. You have to realize what it's used for. It's built for speed, not storage capacity.

      What matters here is the IOPS per $.
      Your 2 TB spinning HD fro $100 might seem cheap, but it only gets some 200 IOPS. So lets say 2 IOPS per dollar.
      These puppies can do some 480k IOPS in a 75/25 read/write workload, which gives us 48 IOPS per dollar.
      So it is in fact 24 times cheaper than a regular HD.
      Then consider electricity, cooling, floor space, floor weight etc.

      --
      -- Another senseless waste of fine bytes.
    13. Re:How much? by maligor · · Score: 1

      Even I have my limits.

      Avnet seems to list something that looks like the PM953 M.2 480GB for 306.24$ per drive, but with a minimum order of 216 units.

      http://avnetexpress.avnet.com/store/em/EMController/Solid-State-Storage/Samsung/MZ1LV480HCHP-00003/_/R-5004524405155/A-5004524405155/An-0?action=part&catalogId=500201&langId=-1&storeId=500201

    14. Re:How much? by Anonymous Coward · · Score: 1

      Hush, don't confuse the consumers, they can only handle one number at a time.
      Storage capacity is the only thing that matters for hard drives. Diagonal inches is the only thing that matters for TV's, brilliant move to switch to widescreen by the way, higher numbers for less screen area!

    15. Re:How much? by fredgiblet · · Score: 2

      The 3D NAND will filter down eventually. It's apparently the way forward as making NAND smaller isn't working anymore.

    16. Re:How much? by smallfries · · Score: 1

      SHUT UP AND TAKE MY m.... Seriously? That much, eh? Maybe I be a little bit later on the adoption curve for this one then.

      --
      Slashdot: where don knuth is an idiot because he cant grasp the awesome power of php
    17. Re:How much? by Christian+Smith · · Score: 3, Insightful

      So $ 9,500 for the 3.84TB one? That's insane.

      This, or a whole rack of short stroked mechanical HDD to achieve the same IOPS performance, and a dedicated runner to keep replacing the failed HDD as you go along, along with the extra power and cooling requirements to boot. Seems like a bargain to me.

    18. Re:How much? by Hognoxious · · Score: 2

      They should really be called shortscreens.

      --
      Confucius say, "Find worm in apple - bad. Find half a worm - worse."
    19. Re:How much? by Kjella · · Score: 1

      IOPS/$ and latency, the time from someone runs a report or clicks a filter until the result is returned means a lot for productivity. Of course you can do a lot with smart systems too, but much like single-thread performance a really fast IO subsystem makes everything easier.

      --
      Live today, because you never know what tomorrow brings
    20. Re:How much? by Anonymous Coward · · Score: 0

      My home lab main server has an Intel 750 and that gave everything a noticeable improvement in performance (I'm running a lot of VMs on an ESX hypervisor). This looks awesome but out of reach for most home users. If I were going to implement a VSAN cluster and wanted high performance between a few hosts these might be the perfect thing. Even if you don't get the chance to have these in your work environment it's nice these things are coming on line because eventually this tech will enter the mainstream consumer market.

    21. Re:How much? by Anonymous Coward · · Score: 0

      For most TV viewing cases, the *height* of the screen determines how big something appears to viewers. That means a 16:9 display needs to be approximately 1.22 times the size (by diagonal measure) of a 4:3 display to maintain the same 'size'.
      For example: A 27" 4:3 display is approximately the same height as a 33" 16:9 display.

      For simplicity's sake, when moving from a 4:3 display to 16:9, you can estimate the 'correct' replacement screen size by adding 20% or 25% to the old number.
      Using the same 27" example, that gives a new screen size of 32.4" to 33.75"

      Math below (in short-hand form)
      4:3 = 12x9 = 144 + 81 = 225 = 15
      16:9 = 16x9 = 256 + 81 = 337 = 18.4
      Size-Factor: 1.2238
      27" @4:3 = 33" @16:9

    22. Re:How much? by OhSoLaMeow · · Score: 1

      Think of the children in Soviet Russia you insensitive clod.

      --
      They can take my LifeAlert pendant when they pry it from my cold dead fingers.
    23. Re:How much? by Anonymous Coward · · Score: 0

      Imagine a beowulf cluster of 4TB SSD

      Imagine a beowulf cluster of ancient redundant slashdot memes, written on scrap lumber and jammed up your ass.

      Jesus that escalated quickly. You insensitive clod.

    24. Re:How much? by Blaskowicz · · Score: 1

      In Soviet Russia, the children think of you.

  2. Hopefully 10,000 RPM models by Anonymous Coward · · Score: 0

    Last SSD I bought was only 7200 RPM, and was disappointing.

    All joking aside, these drives have their uses, and not just the enterprise. The SAS SSD models definitely will be a no-brainer because of SAN demand.

    The PCIe card will come in handy if you are using ESXi, because the hypervisor can use the SSD as swap. Yes, it is definitely slower than RAM... but not devastatingly so as it would be if using a mechanical HDD. So, it will help with overcommits and soften the blow when a user's 2GB VM request winds up being a 128GB appliance for video processing.

    I do wonder if the PCIe card is bootable or not. If it is bootable, it would be useful for a desktop/workstation that doesn't have a M.2 slot.

    1. Re: Hopefully 10,000 RPM models by Anonymous Coward · · Score: 0

      They are ALL bootable. With UEFI 2.3.1 or greater, windows 7 (pro at least I assume). Win8.1 has native drivers (and fast boot:) and of course manufact. has theirs. I'm gettin the 480gb intel 750 pci-e 3.0x4 card. It's ~$380. It should be faster than my evo 840 msata. Coupled with a 12 core 7850K @4.5 we'll see.

  3. figures by Anonymous Coward · · Score: 0

    > Impressive stuff, though it's the sequential read speed that's the real head turner here. At 5,500MB/s (5.5GB/s), Samsung says a user can save a 5GB video file in less than three seconds.

    checks URL, yep more quality hothardware journalism

    1. Re: figures by Anonymous Coward · · Score: 0

      Look, math is hard. The writer picked a number they knew would be safe. Who wants to divide 5 by 5.5 anyways?

    2. Re:figures by pushing-robot · · Score: 1

      save a 5GB file in under 3 seconds

      Read speeds are up to 5.5GBps. Write speeds are up to 1.8GBps.

      Your fast reads are impressive too, but you failed the comprehension benchmark.

      --
      How can I believe you when you tell me what I don't want to hear?
    3. Re:figures by pushing-robot · · Score: 2

      (Mind you, HH still failed on editing, but at least the numbers are right)

      --
      How can I believe you when you tell me what I don't want to hear?
    4. Re:figures by Rockoon · · Score: 1

      save a 5GB file in under 3 seconds

      1.8GB/s x 3s = 5.4GB

      Read speeds are up to 5.5GBps. Write speeds are up to 1.8GBps.
      Your fast reads are impressive too, but you failed the comprehension benchmark.

      fucking dumb much?

      --
      "His name was James Damore."
    5. Re:figures by Anonymous Coward · · Score: 0

      save a 5GB file in under 3 seconds

      1.8GB/s x 3s = 5.4GB

      Read speeds are up to 5.5GBps. Write speeds are up to 1.8GBps.

      Your fast reads are impressive too, but you failed the comprehension benchmark.

      fucking dumb much?

      You are a very sweet and gentle individual.

    6. Re:figures by Anonymous Coward · · Score: 0

      It's possible he missed the subtle change in units when moving from SATA 1/2/3 to SATA Express and M.2. Note he stated speeds in GBps instead of GB/s, which are different by a little over an order of magnitude.

    7. Re:figures by Anonymous Coward · · Score: 0

      err GBps and GB/s are the same? Did I miss something?

    8. Re:figures by Anonymous Coward · · Score: 0

      No, you didn't. It's probably just that the form "GPps" is rarely used, so your parent poster isn't used to seeing it. As written it *is* the same as GB/s, though.

    9. Re:figures by AmiMoJo · · Score: 1

      Are they talking about real, JEDEC standard gigabytes (1,073,741,824 bytes) or those silly new IEC "gigabytes" (1,000,000,000 bytes)?

      --
      const int one = 65536; (Silvermoon, Texture.cs)
      SJW, n: "Someone I don't like, and by the way I'm a fuckwit" - AC
    10. Re:figures by Anonymous Coward · · Score: 0

      Does it matter? Both are vastly inferior to the ones they use in Japan.

    11. Re:figures by Rockoon · · Score: 1

      Note he stated speeds in GBps instead of GB/s, which are different by a little over an order of magnitude.

      Now you are confusing GBps with Gbps.

      --
      "His name was James Damore."
    12. Re:figures by Anonymous Coward · · Score: 0

      JEDEC seems to prefers power of ten metric prefixes, I assume they just don't want to update all their documents. They do always put in a note about power of 2 prefixes just being common use and reference to *bibyte system.

      Power of 2 prefixes are stupid, SI system has been around for far longer with much more application than computer memory.

  4. 1.8 GB/s write. 5 / 1.8 ~ 3 by raymorris · · Score: 1

    It writes at 1.8 GB/s, so the math is correct. The English is bad because they mentioned the read speed in between two mentions of the write speed.

  5. I'm expecting this 6.4TB SSD... by Anonymous Coward · · Score: 0

    To cost more than what I've spent on my last 3 computers combined.

  6. symmetric read/write by slinches · · Score: 1

    Symmetric read/write at 5.5GB/s would be better, but as it is, it would still be quite an upgrade from my current RAID0 array. I'm only getting ~700-800MB/s sequential with four 10krpm SAS drives right now. Of course one of these SSDs will probably cost 5-10x more than a set of those spinning drives. So price/performance really isn't that great of a deal.

    --
    Knowledge Brings Fear
    1. Re:symmetric read/write by Anonymous Coward · · Score: 0

      Symmetric read/write at 5.5GB/s would be better

      For the same price tag, yes.
      In a hypothetical universe where you would select how to distribute read and write speeds when you were ordering the drive I don't think I would distribute them evenly.
      Data that is written but never read can just be dumped to /dev/null. Data that is only read bu never written doesn't exists.
      In practice data is read more than it is written with the exception of data logs.
      When distributing read/write speeds it doesn't seem unreasonable to have a 5:1 ratio in this hypothetical universe.

      In this real world you just want to max out the bus in both directions, that is where the real limit is.

  7. I call BS on the "Update" by Anonymous Coward · · Score: 0

    This article states:

    "As for the PM953, it's an update to the SM951 and is available in M.2 and 2.5-inch form factors at capacities up to 1.92TB"

    There's NO WAY that it's an upgrade to the SM951. The SM951 has been out for only FOUR months now and has only been available to OEMs. Mainstream release to consumers are slated to occur later this year.

  8. price found online for PM953 480GB M.2 by Tumbleweed · · Score: 1

    http://avnetexpress.avnet.com/...

    $306. I don't know if that is wholesale or what.

    1. Re:price found online for PM953 480GB M.2 by Anonymous Coward · · Score: 0

      That is wholesale.

      Minimal order runs in 66k USD

  9. Gamers always get cool shit first! by Anonymous Coward · · Score: 0

    Because gamers are fucking l33333t like, shit, bro. Derp.

  10. Those NORKS Can Make Some After All by Anonymous Coward · · Score: 0

    And you thought Kim Jung Little is all bluster. Seagate? Dump it NOW!

  11. Missing part of the sentence by Anonymous Coward · · Score: 0

    " the latter of which is rated to handle five drive writes per day (32TB) for five years...." ... with a failure rate of what 5%, 10%, 50%, 80%?

    I mean they didn't test a big batch of them for 5 years, since they've only just made them, they tested a batch and some failed after 3 months or 6 months and from that they extrapolated an acceptable fail rate they were prepared to replace for free under warranty.

    Which means a lot of them failed in the test!

    1. Re:Missing part of the sentence by sectokia · · Score: 2

      They don't have to spend months testing it, maybe a couple of days tops. Remember they test with all wareleveling off, a single bit fails extremely quickly (100k writes).

    2. Re:Missing part of the sentence by Anonymous Coward · · Score: 0

      100k writes that's way too high?

      5 years * 365 days & 5 writes a day = 9125 writes to each cell is all they're promising to back with a warranty. Pretty damn crappy but then again the size of them means 370mb per second every second for 5 years is what they're describing.

      So the reliability is more to do with ware-levelling having more cells to write to than anything else. The fact they didn't offer than on the smaller drive suggests it won't last 5 years!

      Not good.

      I don't like SSDs more for the catastrophic failures they produce than the gradual decline. I much prefer RAM (with backup power), its faster and more reliable. So this only comes in if the size of the data is more than 16GB-64GB or so that's workable in RAM now. There aren't many data sets that match that.

  12. early MLC endurance? :o by citizenr · · Score: 1

    > 6.4TB, the latter of which is rated to handle five drive writes per day (32TB) for five years

    ~10K write cycles. This sounds rather good.

    --
    Who logs in to gdm? Not I, said the duck.
    1. Re:early MLC endurance? :o by Anonymous Coward · · Score: 0

      ~10K write cycles. This sounds rather good.

      Sound rather average. 10k write cycles is standard for flash memories. When designing a system, before you have selected what memories to use, you can guesstimate it to 10k write cycles.

      Unless you meant Kelvin. Then it sound really crappy. A device that requires that kind of cooling for writes seems impractical.

  13. Slapping together vs enterprise storage. $1200 car by Anonymous Coward · · Score: 1

    Slapping together a bare bones system costs a lot less than buying enterprise storage. I've done that recently, and spent about $1200, like you said. I bought a use raid card on eBay because the card cost almost $1200 new. Then you get a proper chassis with reliable, hotswappable cooling for all those drives running 24/7, redundant power supplies, a backplane, etc. About the only way you're going to get enterprise grade raid for $1200 is to buy used, which I often do.

  14. Proofread, idiots! by Anonymous Coward · · Score: 0

    "The fastest of bunch"

  15. Memories by Artem+S.+Tashkinov · · Score: 2

    I remember just ten years ago RAM, yeah RAM, was slower than these drives. Time flies!

  16. TRIM Support? by Anonymous Coward · · Score: 0

    But do they properly support the TRIM command?

  17. Ah: Some REAL "news for nerds" @ last! apk by Anonymous Coward · · Score: 0

    See subject: It's what makes being a nerd, great... kudos to the nerds doing the inventing here! It's truly, inspiring...

    * :)

    QUOTING THE SHOW "The 6 Million Dollar Man" (for those of you old enough to know that show that is):

    "We can make him BETTER than he was before - Better, Faster, Stronger..."

    (... & It's far better + more useful news than "Gay Marriage" carrots tossed out there, that are intended to distract you from REAL issues out there, or good news like this...)

    That's what this does for SSD, the "wave of the future" (what I've been using @ HOME no less, ones based on "REAL RAM", not flash, since oh, 2000 or thereabouts...) - & THAT future?

    IS, truly, NOW!

    APK

    P.S.=> I wonder (since I haven't read the article yet) - IS IT BASED ON/INTENDED for Pci-Ex architecture too? That is a HUGE help to SSD's, overcoming bus speed limitations... apk