Intel expands DC S3500 SSD family with monster 2.5” units, mini M.2 models

Intel has been making server SSDs since the X25-E debuted six years ago. Those drives have apparently grown quite popular, too, because the company claims it’s been the number-one provider of SATA SSDs to datacenters for the past three years. In a bid to keep that streak going, Intel is introducing much larger—and much smaller—entries in the DC S3500 Series that arrived last year.

Arguably the biggest news is the addition of 1.2TB and 1.6TB capacities to the 2.5″ lineup. That family previously topped out at 800GB, but a new spin of the proprietery Intel controller allows the S3500 to address twice as much flash. The controller tweaks are limited to supporting more NAND; Intel says the updated chip has the same eight-channel architecture and SATA 6Gbps interface as its forebear.

Here’s how the new 2.5-inchers stack up against the existing members of the family:

Capacity Max sequential (MB/s) 4KB random (IOps) Endurance Price
Read Write Read Write
1.6TB 500 460 65,500 18,500 880TB $1444
1.2TB 500 460 65,000 14,600 660TB $1099
800GB 500 450 75,000 11,500 450TB $740
600GB 500 410 75,000 11,000 330TB $450
480GB 500 410 75,000 11,000 275TB $400
300GB 500 315 75,000 9,000 170TB $230
240GB 500 260 75,000 7,500 140TB $210
160GB 500 175 75,000 7,500 100TB $160
120GB 475 135 75,000 4,600 70TB $110
80GB 445 100 70,000 7,000 45TB $100

According to the official specs, the larger S3500s are slightly slower than their siblings with random reads but faster with random writes. They have higher endurance ratings, too, though only in terms of total bytes written. All the members of the DC S3500 Series are rated for 0.3 full drives writes per day for the length of their five-year warranty.

Source: Intel

The smaller additions to the S3500 family comprise M.2 models with 80GB, 120GB, and 340GB capacities. These mini SSDs have the same controller and SATA interface as the 2.5-inchers, but they come on a much smaller M.2 2280 “gumstick” form factor. They’re meant to be used as boot drives and in embedded systems, microservers, and digital signage.

Cue specs:

Capacity Max sequential (MB/s) 4KB random (IOps) Endurance Price
Read Write Read Write
340GB 480 355 67,000 14,500 180TB $314
120GB 440 160 67,000 12,000 70TB $124
80GB 340 110 67,000 8,300 45TB $99

The M.2 units have lower random read rates but higher random write speeds than their 2.5″ counterparts. Sound familiar? There are also some differences in sequential performance, all in favor of the mini drives.

Like the rest of the DC S3500 Series, the M.2 variants are built with Intel’s own 20-nm MLC NAND. Other common features include 256-bit AES encryption hardware, end-to-end data protection, and capacitor-based power-loss protection. Server-grade validation testing? Check.

Given their enterprise credentials, the new DC S3500 SSDs aren’t too expensive. Just about all of ’em ring in at under $1/GB, which probably means someone needs a phone call. The drives are due to be available in the distribution channel, so expect to see them popping up at places like Newegg before long.

Comments closed
    • Scrotos
    • 5 years ago

    Given their enterprise credentials, you’d think there would be some SAS offerings.

      • gospadin
      • 5 years ago

      [url<]http://www.hgst.com/press-room/press-releases/HGST-Extends-Development-Agreement-with-Intel-for-HGST-s-Enterprise-Class-SAS-SSDs[/url<] Intel and HGST are partners in some of the HGST SAS SSDs.

    • Waco
    • 5 years ago

    Anyone else wonder why the endurance specs are so low? We’ve seen time and time again that NAND Flash can handle anywhere from 1000-3000 write/erase cycles for MLC…yet we’re still seeing drives quoted with barely over 500 total drive writes.

    I assume it’s because of write amplification, but still, I know they can measure that. Why not have a reasonable number based on total erase cycles?

      • just brew it!
      • 5 years ago

      How are total erase cycles any more meaningful to the end user? The spec is likely based on a worst-case (or nearly so) scenario for write amplification; if you’re using TRIM and/or writing a lot of sequential data I expect actual endurance will be much better.

        • gospadin
        • 5 years ago

        This. Additionally, “drive writes per day” is ultimately a much more friendly metric for customers than providing an arbitrary number that is so impacted by workload choice and internal management algorithms.

    • UberGerbil
    • 5 years ago

    340GB? Can’t say I’ve seen that particular capacity before.

      • just brew it!
      • 5 years ago

      Me either. But given the variable amount of over-provisioning and the ability of flash controllers to utilize fewer flash channels than their maximum capability, SSD capacities are basically at the arbitrary whim of the manufacturer.

    • TwoEars
    • 5 years ago

    I’m liking that 1.6 TB model a lot.

    I have some clients with serious workstations data processing demands and 8 of those in raid-5 with a LSI 9361-8i controller should be just the ticket.

    • hubick
    • 5 years ago

    I want a fast PCIEx4 M.2 to use that slot on my new X99 board.

      • UnfriendlyFire
      • 5 years ago

      I’m more interested in consumer laptops having the M.2 slots in the future, and more importantly, M.2 SSDs’ price per GB come down to the point where it’s cost competitive against 2.5″ SSDs.

    • xonarlover
    • 5 years ago

    120 TB for $124? I gotta get two for my RAID-0 N(S)A(S) array. 🙂 😛

    [url<]http://i.imgur.com/HyrzIC7.png[/url<] EDIT: Well, schucks. It seems intel reduced the storage capacity by a factor of 1,000 with a firmware update. 🙁 Now how am I going to explain the purchase of 2,000 SSD's to my boss, Arati Prabhakar. 🙁

      • Dissonance
      • 5 years ago

      Fixed.

      • Hattig
      • 5 years ago

      Hehe, its write endurance is lower than its capacity!

        • Wirko
        • 5 years ago

        That’s normal. What we’re dealing with here is MMLC, or More and More Levels Cell, the only suitable tech to make a $.001&GB SSD. The theoretical maximum is estimated at 2000 bits per cell. However, cells are usually full earlier and stop accepting new bits.

        Note that this is achieved by using smaller electrons, and Intel has some headstart here compared to others.

          • xonarlover
          • 5 years ago

          I heard intel was working on using quarks instead of electrons… their atoms were causing too great of a hassle.

          EDIT: Well, shoot. Now intel’s going to push back the release of their quark ssd’s and NAND to Q4 2080 from Q2 2060. 🙁 I guess I’ll have to settle with Micron’s 5 nm phase change SSD. 😛

          I miss the good old days of the random HDD lottery. If only China hadn’t nuked Thailand in WWIII, then we could’ve had a storage solution we didn’t have to swap every two months.

            • UnfriendlyFire
            • 5 years ago

            TR administrators, did you check if the server time was correct? Something’s not right here.

          • Generic
          • 5 years ago

          Smaller electrons?

          You have my full attention sir.

          Edit:
          Ah shucks. Xonarlover beat me by a mile. >.<

Pin It on Pinterest

Share This