Intel reveals a whole arsenal of enterprise hardware at its Data-Centric Innovation Day

In case you haven't heard, Intel hosted an event for the press today that it called "Data-Centric Innovation Day." The primary purpose of the event was to promote a bunch of new products: the Agilex FPGAs, new SSDs, new Ethernet adapters, and most critically, the Cascade Lake family of Xeon CPUs. We weren't privy to any press materials before the presentation, so this is just a quick overview of all the new hardware on display today.

Obviously, the biggest news of the day is that Intel has launched its second-generation Xeon Scalable processors. These chips are the successors to the Skylake-based platform launched in 2017, and they were supposed to arrive at the end of last year. That didn't happen, but the chips are out now, and according to Intel you should be able to order machine with one as I write this.

The new Xeons haven't showed up on ARK yet, but Intel says there are "over 50" chips in the family. The company was eager to point out during its presentation that that number does not include some portion of bespoke processors. The chips range from single-socket eight-core models all the way up to a 400-W, 56-core behemoth with two CPU dice and a full twelve memory channels. All of the new chips are based on a revised CPU core that includes "DLBoost" instructions to accelerate AI inferencing as well as hardware mitigations for most Spectre variants.

Navin Shenoy holds up a stick of Optane DC persistent memory.

Most of the new chips—but not all—also support Optane persistent memory. That sounds exactly like what it is: regular old DIMMs, but with Optane memory onboard instead of DDR4. As long as the machine has at least one module of DDR4 DRAM, the rest of its slots can be packed with extremely high-density Optane DIMMs. Intel says that this enables up to 4.5 TB of memory per socket, which is certainly extremely impressive, though we're curious to see what the performance implications of such a configuration would be once software is written to take advantage of it.

Besides the big Xeons, Intel briefly showed the next series of Xeon D processors. Like those that came before, the new Xeon D-1600 family uses the same full-fat CPU cores as its heavyweight cousins, but stuffs them into an SoC-style package that Intel says can go as low as 27W. These chips are primarily intended for network applications, and as such integrate Intel's QuickAssist technology for hardware-accelerated encryption. Unfortunately, that's about all Intel had to say on these chips.

Other than Xeons, Intel very quickly announced a fair few non-CPU components. On the storage side of things, we saw a version of the Optane SSD DC D4800X with dual U.2 ports. That feature isn't designed to improve performance, but instead to increase availability. The company also showed the SSD D5-P4326 in EDSFF "ruler" form factor. These drives use QLC NAND flash to cram nearly 31 TB into a stick roughly the size of a typical classroom straight-edge. Intel says you could use these drives to stuff around a petabyte of storage in a 1U enclosure.

On the network side of things, Intel announced the 800-Series Ethernet adapters, codenamed Columbiaville. The add-in cards Intel showed hook up to 100-gigabit Ethernet, which is the sort of transfer rate that us mere mortals can only imagine. Intel talked about several technologies that these controllers offer: Application Design Queues are a form of hardware quality-of-service assurance, while Dynamic Device Personalization—carried over from the 700-series adapters—allows extensive reconfiguration of the adapter without resetting the system.

Finally, Intel showed off its first-ever FPGAs developed and built right at home: the Agilex series. It's an all-new brand for the company, and it describes a series of 10-nm FPGAs that use Embedded Multi-die Interconnect Bridges (EMIBs) to hook up to chiplets offering various I/O and other functionality. These are the only products that Intel announced today which won't be coming to market anytime soon; the very first samples of Agilex will be available in Q3 of this year. Intel claims that these chips will deliver up to 40 TFLOPs of DSP performance.

Comments closed
    • GTVic
    • 5 months ago

    Is there a successor to the W-Series in this set of CPUs?

    • Mr Bill
    • 5 months ago

    I suppose, symmetrical pairing of optane modules across all available memory channels except one, would be the optimal layout?

      • DavidC1
      • 5 months ago

      I’m not sure what you mean by “except one”?

      You can have 6 DRAM DIMMs and 6 Optane PMMs per socket. Some amount of DRAM is necessary, so they’ll essentially be paired.

        • Mr Bill
        • 5 months ago

        [quote<]That sounds exactly like what it is: regular old DIMMs, but with Optane memory onboard instead of DDR4. [b<]As long as the machine has at least one module of DDR4 DRAM, [/b<]the rest of its slots can be packed with extremely high-density Optane DIMMs. [/quote<] I was supposing you might want to populate at least one complete memory channel with DDR4 DRAM so as to get best performance from that channel (e.g. 2 or 3 sticks of DDR4 DRAM) Instead of just having a single stick of DDR4 DRAM on one of the memory slots.

          • DavidC1
          • 5 months ago

          I think its the TR article that’s confusing.

          It should say “As long as every memory channel has at least one module of DDR4 DRAM”.

          It can support 2 DIMMs per channel, so a 6 channel system has 12 DIMM slots.

            • Waco
            • 5 months ago

            That’s right. Per Dell’s configuration tool (fastest way to easily check) these are the requirements if selecting NVDIMMs:

            Performance Optimized with NVDIMM requires specific amount of Memory DIMMs with 2 Processors configuration. • 1 x NVDIMM, 12 x 16GB/32GB RDIMM • 1 x NVDIMM, 23 x 32GB RDIMM • 2 x NVDIMM, 12 x 16GB/32GB RDIMM • 2 x NVDIMM, 22 x 32GB RDIMM • 4 x NVDIMM, 12 x 16GB/32GB RDIMM • 4 x NVDIMM, 20 x 32GB RDIMM • 6 x NVDIMM, 12 x 16GB/32GB RDIMM • 6 x NVDIMM, 18 x 32GB RDIMM • 12 x NVDIMM, 12 x 16GB/32GB RDIMM

    • freebird
    • 5 months ago

    Is this news why AMD’s stock was up about $3 from yesterday morning to today???

    • Sahrin
    • 5 months ago

    Still desperately trying to get back some of the $17B you blew on the Altera acquisition, eh Intel?

    I like how they advertise “>50 SKU’s” like it’s a good thing. “Our product lineup is so complex, that your acquisition’s team won’t even have time to evaluate the competition!”

    Think I may have just put my finger on the trigger.

      • chuckula
      • 5 months ago

      Considering Intel’s programmable systems group brought in about 50% more revenue last quarter than AMD’s entire “enterprise” business combined with all their game console revenue, I’m pretty sure that Lisa Su would gladly sell your organs on the black market to get AMD into an equally “desperate” situation.

        • Sahrin
        • 5 months ago

        There’s only one person here claiming Intel’s situation is desperate.

        I love you Fox News inspired trolls.

        “Intel wasted money buying altera.”

        “Lol, no, Intel isn’t on the verge of bankruptcy and dying from cholera moron!”

        Please tell me you have more to contribute than this.

          • chuckula
          • 5 months ago

          [quote<] xThere’s only one person here claiming Intel’s situation is desperate[/quote<] Yes, and that would be you in the first sentence of your idiotic sh1tpost. As for the rest of it, I'll proudly admit to being a smartass on occasion. Operative word there being "smart". Your problem seems to be that all of your posts are made from the point of view of a dumbass. Operative word being dumb.

          • Mr Bill
          • 5 months ago

          You can call him a shill. But you can’t call him an ignorant shill.

      • DavidC1
      • 5 months ago

      I saw a reply on the article that says the lineup doesn’t mesh with what he needs and is expecting more SKUs.

      They have so many because customers ask for them.

    • jarder
    • 5 months ago

    So did Intel kill off their Avoton line of Atom-based server chips? the Internet’s not giving me a straight answer.

    I know they ran into some hardware problems 2 years ago with c2000 based boards dying and such:
    [url<]https://www.extremetech.com/computing/244074-intel-atom-c2000-bug-killing-products-multiple-manufacturers[/url<] [url<]https://www.reddit.com/r/homelab/comments/8vhsle/asrock_c2750d4i_died_looking_for_replacement/[/url<] But the C3000 series boards, after showing up in a few reviews last year, don't seem to be available. I suppose with the lower-power Xeon-d chips, Avoton's niche has gotten smaller, but there was still a case for them IMO.

      • chuckula
      • 5 months ago

      They didn’t “kill off” Avaton any more than they “killed off” Broadwell. Avoton was just an older product line that’s been updated by Denverton and will certainly be updated again in the future with Tremont core Atoms.

      As for the C3000 series “not being available,” even Newegg — which isn’t the best place to look for server hardware by any margin — has plenty of products available: [url<]https://m.newegg.com/ProductList/keywordsearch?keyword=Atom+C3000[/url<]

        • Anonymous Coward
        • 5 months ago

        I think they’d be foolish to drop “tiny core” server products at this point, there seems to be some inertia behind using ARM in the server space, so apparently someone with money believes in the concept of small cores. They need to keep them around at least to do battle against ARM server products, regardless of profit margin, IMO.

          • ronch
          • 5 months ago

          What about AMD’s Opteron A1100 series based on [s<]small[/s<] A57 ARM cores? It's probably putrifying already, isn't it? Edit - oops looks like it's stuck between the land of the dead and the land of the living. There's still [url=http://www.lenovator.com/product/103.html#params<]someone[/url<] selling stuff based on it but it's not exactly HP or Cray. I wonder who else is selling them, and how many a month.. sorry too lazy to bother looking.

            • Anonymous Coward
            • 5 months ago

            I’m using AWS servers based on ARM, apparently at the beginning of that effort AWS and AMD worked together, but they parted ways. I can’t comment on the economics from the perspective on Amazon, but the servers work fine, sometimes they represent the optimal value for “renters”.

            The cores are weak but a lot of bandwidth is available, as are a fair number of cores.

        • jarder
        • 5 months ago

        Good point, I was basing things off the previous code-name. Looking into your list, of the list of 7 systems you point to, only 3 seem to exist, and one of those is out of stock, then, the two remaining are both based on the 8-core C3758. I was mostly interested in the 16-core variant, and that’s the one that most reviews looked at e.g.:
        [url<]https://www.servethehome.com/intel-atom-c3955-16-core-top-end-linux-benchmarks-review/[/url<] I did find one based on the 16-core chip on newegg: [url<]https://www.newegg.com/Product/Product.aspx?Item=9SIAH9B8Y46428&Description=c3955&cm_re=c3955-_-9SIAH9B8Y46428-_-Product[/url<] but that's rather expensive at $1272. Based on the ARK pricing (Yes, I know), it should be only about $250 more than the 8-core board, not $800 more, something's not right there. Looking at newegg some more, it looks like the reports of the C2000 series failure problems might have been under-reported, check out the reviews on: [url<]https://www.newegg.com/Product/Product.aspx?item=N82E16813157475[/url<] The more recent reviews are particularly damning. That's giving me second thoughts about my potential choices for a new NAS/server build.

          • Godel
          • 5 months ago

          The third anonymous Newegg review says the bug is fixed with the latest revision.

          “If it says (C0) on the box, this means it is NOT affected by the C2000 Clock death bug.
          The fixed model number on the board I received back from AsRock via RMA is: C2750D4I/C0.
          The one it replaced (and the one I purchased from Newegg) was: C2750D4I/M/ASRK.

          The fixed board itself is silkscreened with hardware revision 1.03.”

          But the reviews all suggest that service from the company if your board does fail is lousy.

    • Anonymous Coward
    • 5 months ago

    I’ve read other places that the newly launched Xeon D-1600 chips are still Broadwell cores, but with refined manufacturing. Thats a very curious choice by Intel if its accurate.

      • MOSFET
      • 5 months ago

      Interesting, and thanks for pointing that out.

      I would guess that the 1600 series stays fairly low power like the 1500 series. If that’s the case, staying with Broadwell probably makes sense. Maybe FIVR really is superior at the decidedly middle-of-the-road TDPs that the 1500 series hit. As long as the chip has middling TDP, just enough clock speed for decent compute, and great networking it should work. Clock for clock with Skylake, Broadwell should be awfully close, and it’s probably all about power consumption anyway. The 2100 series didn’t seem to have the same warm reception as the 1500 initially did.

      Notice I use the words guess, maybe, and probably (more than once). 😉

        • Anonymous Coward
        • 5 months ago

        I liked the theory that Ryzen (as we currently know it) is fundamentally a simpler and for many purposes more efficient core than Intel *lake cores, would be interesting if Intel had a similar thought about their *well cores.

    • ronch
    • 5 months ago

    Look at the die shot on the first slide there. Do they look a little different from past Intel cores? Seems to me they changed quite a lot of plumbing in there. Looks basically the same but there are new structures and all. Or is this just an effect of moving to 10nm, like how the K8 core looked squarish (instead of rectangular) when it was built on 65nm (but then became rectangular again at 45nm).

      • DavidC1
      • 5 months ago

      The structure seems to be the same to me. Go find skylake server and find the XCC die. 2nd gen Xeon Scalable is on 14nm so process didn’t change either.

      The biggest differences are the hardware fixes to some Meltdown/Spectre and Optane DIMM support. The former should take minimal amount of space to be noticeable and the latter was supposedly disabled in Skylake version of Xeon Scalable, so I doubt you would notice it either.

        • mczak
        • 5 months ago

        They also support a couple new instructions (AVX-512 VNNI). But you wouldn’t see that on die shots neither (should only require very minor changes in the simd unit, probably about as minor as what’s required for Meltdown/Spectre).

    • ronch
    • 5 months ago

    This is epic news. <cough cough!>

    • Anonymous Coward
    • 5 months ago

    I wonder how the economics of 1 vs 2 socket servers will change with this. Dual socket has been the sweet spot for a good chunk of time now, but these are getting to be some huge, complicated sockets.

    It will also be interesting to see what possibilities pop up for that Optane memory.

    • chuckula
    • 5 months ago

    Tsk tsk.
    Caught Intel in a lie on the first slide.
    Xeons can scale way past 8 sockets with glueless interconnects. They should have noted that “1 to 8 sockets” is just using the vanilla ultrapath links.

    Case in point: [url<]https://www.anandtech.com/show/13522/896-xeon-cores-in-one-pc-microsofts-new-x86-datacenter-class-machines-running-windows[/url<] Oh and that's from last year.

      • Anonymous Coward
      • 5 months ago

      Yikes, a bit of Google and indeed Intel mentions 32-socket systems:

      [quote<]Configure native 4- and 8-socket CPU configurations and extend to up to 32-socket systems via third-party OEM node controllers.[/quote<] Who uses such a thing, and for what purpose?

        • Neutronbeam
        • 5 months ago

        We CAN’T tell you that… Just ignore any time continuum flux incidents you may or may not encounter.

          • Anonymous Coward
          • 5 months ago

          I do a certain amount of processing at scale myself, nothing too much to brag about, but I’m wondering what niche they have there in which 32 sockets in one unified memory space makes more sense that 32 sockets on 16 or even 32 boxes. I’ve seen 4 sockets as a means to get more memory. But mostly it seems pretty clever to structure software so it can be distributed in reasonably small chunks, not only saving money on the server, but also making the chunks accessible individually in case of errors (so only part of the big system has failed, and that part can ideally be fixed and rerun).

            • Oem
            • 4 months ago

            Remember the Intel 286? “Structuring software” isn’t cheap, in fact in time and money developing complex software is one of the most expensive parts of business, government and research projects that have a software component. The 286 had a segmented memory architecture that made developing software for it a PITA, far more time and trouble to build software that needed more than 64K of data or code in one conceptually contiguous space than it should have been. The 386 finally introduced a flat memory model, and not coincidentally enabled Windows NT & Windows 95 and a hypergrowth phase of the PC ecosystem. From a software developer point of view, if you have a problem that needs X amount of memory, coding it on an architecture that allows >= X amount of memory in a single shared space can be vastly simpler than adding a layer of complexity to deal with architectural limitations. This development time and cognitive load has nothing to do with the problem you are actually trying to solve, and everything to do with implementation details and limitations. If the problem is complex you may have your mental hands full just trying to understand it and your implementation. The “segmented architecture” approach may not only add time and / or cost but perhaps it may even increase the possibility that there are subtle undiscovered bugs in your model by obscuring the implementation and spreading the understanding of the codebase across more minds than strictly necessary on an “ideal” flat-model implementation. For certain deep-pocketed customers, providing a “clean” programming model to their developer team may be worth ponying up the extra dough for such an extreme system.

      • Mr Bill
      • 4 months ago

      That has got to be running the most expensive copy of windows, ever.

    • Krogoth
    • 5 months ago

    Intel’s secret weapon has been unleashed.

    SUPER SHILL GLUE!

    Now you will witness the full shilling power of this multi-packaged chip.

      • jihadjoe
      • 5 months ago

      Two shills one cap (heatspreader)?

        • Mr Bill
        • 5 months ago

        With multiple shillers, who needs a heat spreader?

Pin It on Pinterest

Share This