Stories
Slash Boxes
Comments

SoylentNews is people

posted by Fnord666 on Wednesday February 07 2018, @06:13PM   Printer-friendly
from the do-they-have-barbeque-flavor? dept.

Submitted via IRC for TheMightyBuzzard

Ampere, a new chip company run by former Intel president Renee James, came out of stealth today with a brand-new highly efficient Arm-based server chip targeted at hyperscale data centers.

The company's first chip is a custom core Armv8-A 64-bit server operating at up to 3.3 GHz with 1TB of memory at a power envelope of 125 watts. Although James was not ready to share pricing, she promised that the chip would offer unsurpassed price/performance that would exceed any high performance computing chip out there.

The company has a couple of other products in the works as well, which it will unveil in the future.

Source: TechCrunch


Original Submission

This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
(1)
  • (Score: 3, Interesting) by takyon on Wednesday February 07 2018, @06:52PM (6 children)

    by takyon (881) Subscriber Badge <{takyon} {at} {soylentnews.org}> on Wednesday February 07 2018, @06:52PM (#634491) Journal

    What's the core count? How does it beat Xeons, Epyc, and other ARM server chips (such as the 48-core Qualcomm Centriq)?

    The story also implies that this unreleased ARM chip is affected by Spectre:

    Her company was designing this chip and getting ready to unveil it to the world when the Spectre and Meltdown bug news hit last month. As James acknowledged, no modern chip that uses speculative execution can escape these bugs, but she said the Arm exposure was minimal and the chips will have the patches built into them when they hit the market later this year.

    --
    [SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
    • (Score: -1, Spam) by Anonymous Coward on Wednesday February 07 2018, @08:10PM

      by Anonymous Coward on Wednesday February 07 2018, @08:10PM (#634541)

      She screamed. Oh, she screamed. In response, the sound of little boys cheering was heard. Chairs. A tile floor. A chalk board. It was a classroom.

      The man was vigorously moving his hips and slamming his fist into the woman's face. It might be more accurate to say that he could stop neither his fists nor his hips. The woman screamed for help. However, the children only cheered. Rather than caring about the woman's well-being, it would be more accurate to say that the boys were actively cheering for her demise.

      Every time the fist collided with the woman, mankind took a microscopic step forward towards a future where men's rights were respected. The children knew this, which is why they were so excited. A bright future awaited them.

      When silence finally descended upon the woman, the children could no longer contain their excitement. Endless cheers and clapping were heard from within the classroom. The woman's motion had been completely replaced by the children's desire for freedom.

    • (Score: 2) by DannyB on Wednesday February 07 2018, @08:27PM (4 children)

      by DannyB (5839) Subscriber Badge on Wednesday February 07 2018, @08:27PM (#634549)

      It doesn't necessarily have to beat Xeons. It just needs to have reasonable performance, and a lower cost to achieve the same level of performance as the Xeon. The foregoing statement only holds true for applications where you have LOTS of chips1.

      It probably does need to beat Centriq, or come close. Again, cost, both initial capital, and operational, and floor space square meter costs do matter.

      1eg, not a desktop, where it simply is not acceptable to replace one chip with, say two chips, or 8 cores with 16 cores. Single thread performance matters. But not as much for large cluster applications where you simply tell Kubernetes to add a few hundred extra nodes to service your workload.

      • (Score: 2) by TheRaven on Thursday February 08 2018, @09:49AM (3 children)

        by TheRaven (270) on Thursday February 08 2018, @09:49AM (#634797) Journal
        For cloud workloads, more slower cores may be better. The easiest way to stop speculative execution vulnerabilities mattering is to have one core per security domain. If you can host VMs that support hot-plug CPUs and can dynamically increase the number of cores that they have access to, rather than the number of time slices on a small number, then you can still do elastic scaling and not worry too much about side channels (cache ones still exist, but they're difficult to exploit if the cache is the only side channel, and can be mitigated with static cache provisioning).
        --
        sudo mod me up
        • (Score: 2) by DannyB on Thursday February 08 2018, @03:11PM

          by DannyB (5839) Subscriber Badge on Thursday February 08 2018, @03:11PM (#634945)

          one core per security domain

          Interesting idea. Never thought of that.

          Maybe have one socket (with one or more cores) per security domain.

          The idea being that even if you can Spectre / Meltdown to peek kernel memory, you can only learn secrets related to the security domain your attack code is executing in. On a multi-tenant cloud system, you can't learn secrets about other customers. Or on a Google like system, you might have successfully attacked, say, blogger nodes, but you wouldn't ever see processes from say, YouTube, or Gmail, to contrive an example.

        • (Score: 2) by frojack on Thursday February 08 2018, @09:20PM (1 child)

          by frojack (1554) Subscriber Badge on Thursday February 08 2018, @09:20PM (#635210) Journal

          It might be easier to stop speculative execution by simply not building it into the processor in the first place.

          I'd like to see what percentage of typical job time is saved by speculative execution.

          If it were all that great why not build that functionality into the compilers, and spend an extra two minutes in optimization at compile time and avoid the risk?

          If its sot significant, just figure out how much faster the clock speed needs to be to make up for it.

          --
          No, you are mistaken. I've always had this sig.
          • (Score: 2) by TheRaven on Monday February 12 2018, @12:35PM

            by TheRaven (270) on Monday February 12 2018, @12:35PM (#636661) Journal

            I'd like to see what percentage of typical job time is saved by speculative execution.

            On a modern Intel chip, you have up to around 180 instructions in flight at a time. The typical heuristic is that you have, on average, a branch every 7 instructions. Every instruction between the branch being issued and the instruction before it that provides the branch condition reaching writeback is speculative. This means that, on average, around 96% of your instructions are speculatively executed.

            On simpler pipelines, the number is a lot lower. A simple 7-state in-order pipeline is only speculatively executing around 50% of its instructions. So, if you disable speculative instructions entirely then you'll take a 50% performance hit on simple (read: slow) pipelines or around a 96% performance hit on high-end pipelines in the worst case. It isn't quite that bad in the average case, because (as these vulnerabilities showed) speculative execution isn't perfect, so you won't see a difference between not doing speculation and the cases where you'd see incorrect speculation. I'd expect that on a simple in-order core you'd only see around a 30% performance hit and on a high-end Intel core around an 80% hit.

            That said, we only do speculative execution because most code is written in languages like C that don't provide enough high-level parallelism to keep a CPU busy. If you were to design a CPU to run a language with an abstract machine like Erlang, then you could get away without speculative execution by running instructions from another thread.

            If it were all that great why not build that functionality into the compilers, and spend an extra two minutes in optimization at compile time and avoid the risk?

            If the compiler could statically determine branch targets, then it wouldn't bother inserting branches. You can do the classical GPU approach and execute both branches and then discard the results that you don't want, but then you end up seeing performance drop by 50% for each conditional branch.

            If its sot significant, just figure out how much faster the clock speed needs to be to make up for it.

            Faster than you can build, and a lot faster than you can cool (I forget the exact relationship between power consumption and clock rate, it's either square or cube - this is why hardly anything runs at over 2GHz). For a modern Intel chip to reach the same performance without speculative execution, you'd need to go around 10-20GHz, which no one has come close to being able to build (at least, not in anything that didn't run briefly with liquid nitrogen poured on it before burning out).

            --
            sudo mod me up
  • (Score: 2) by richtopia on Wednesday February 07 2018, @07:01PM

    by richtopia (3160) Subscriber Badge on Wednesday February 07 2018, @07:01PM (#634492) Homepage Journal

    I am reminded of Qualcomm's new ARM chip for the datacenter, the Centriq 2400: https://www.anandtech.com/show/12025/qualcomm-launches-48core-centriq-for-1995-arm-servers-for-cloud-native-applications [anandtech.com]

    However, the Centriq is 48 cores at 2.6Ghz, while this Ampere chip is 3.3Ghz with no comment on number of cores. Perhaps they are targeting different market segments.

  • (Score: 1, Interesting) by Anonymous Coward on Wednesday February 07 2018, @07:45PM

    by Anonymous Coward on Wednesday February 07 2018, @07:45PM (#634516)

    Would like to know shopping list:

    Single thread/core performance relative to other options
    Number of cores, cache sizes, multicore and memory performance.
    Status on current spec exec security issues
    Mystical marketing, or open access to the actual hardware extra firmware story
    Cost, including ease of integration into a motherboard
    Attitude towards errata sheet. (Significance of bugs and openness of where they are.)

  • (Score: 2, Informative) by Anonymous Coward on Wednesday February 07 2018, @08:01PM (2 children)

    by Anonymous Coward on Wednesday February 07 2018, @08:01PM (#634531)

    oh great, just what the world needs. more slaveware.

    • (Score: 1, Insightful) by Anonymous Coward on Wednesday February 07 2018, @08:15PM (1 child)

      by Anonymous Coward on Wednesday February 07 2018, @08:15PM (#634543)

      For now, it's going to take Big Corps and the hordes of consumers to fund the development of advanced computing technology that will ultimately become available even to the individual.

      • (Score: 0) by Anonymous Coward on Thursday February 08 2018, @07:18PM

        by Anonymous Coward on Thursday February 08 2018, @07:18PM (#635135)

        that's just apathy and boot licking disguised as wisdom.

  • (Score: 4, Interesting) by requerdanos on Wednesday February 07 2018, @08:08PM (14 children)

    by requerdanos (5997) Subscriber Badge on Wednesday February 07 2018, @08:08PM (#634537) Journal

    The company's first chip is a custom core Armv8-A 64-bit server operating at up to 3.3 GHz

    I have a Friendly Arm NanoPC-T3, which is based around an 8-core Armv8 64-bit operating at up to 1.4 GHz. I have "FriendlyCore aarch64" (which is based on ubuntu) installed on it. I would tell you how fast it is, but the word "fast" does not apply here; I would have to tell you how slow it is. And the answer is pretty darned slow.

    So, it concerns me when an apparently similar chip, albeit with higher clock speed ("2.4 x as fast?"), is touted as having qualities such as "unsurpassed price/performance that would exceed any high performance computing chip out there."

    Unless they have made some quantum leaps, and I sincerely hope they have; I would welcome a unit to review (hint hint), it would have to be almost free to have better price/performance than a ho-hum average AMD FX/Zen or Intel Xeon.

    • (Score: 5, Insightful) by edIII on Wednesday February 07 2018, @08:18PM (6 children)

      by edIII (791) Subscriber Badge on Wednesday February 07 2018, @08:18PM (#634544)

      I have no interest in price/performance anymore. Whatever it is, it is. I'm extremely interested in price/security/openness. How open is the architecture? Are there any blobs anywhere? Is there a security chip like the Intel ME? If so, do I have access to the source code? Can I compile my own security engine to run on this dedicated security processor?

      Those are the questions I have now. I would build a system today with far less power than the bleeding edge processors out there, but all the security we want and need. I firmly believe now that security can ONLY be obtained with full and absolute transparency. No security is obtained through obscurity of the methods and processes.

      Intel and AMD can both do whatever the hell they want, but the first company to deliver the security we truly need will start getting a lot of orders. Even if they're lower on the performance/feature totem pole.

      • (Score: 3, Interesting) by requerdanos on Wednesday February 07 2018, @08:40PM (1 child)

        by requerdanos (5997) Subscriber Badge on Wednesday February 07 2018, @08:40PM (#634554) Journal

        security can ONLY be obtained with full and absolute transparency.

        This is a truism; I think the love of "management engine" type controls is of control-freak large IT departments in large organizations.

        If I wanted to reboot remote computers, I would install a small singleboard computer that could electronically press the reset button. If I wanted to change their bios remotely, I don't know what I would do beyond hiring someone to trudge over there and poke through the BIOS, admittedly, but a technical solution I'd approve would definitely not involve malware buried deeper even than your average rootkit.

        As usual, my requirements and desires do not sync with those of the market at large. And this time, even simple, proven knowledge like "no security in obscurity" isn't standing in the way.

        I have no interest in price/performance anymore. Whatever it is, it is. I'm extremely interested in price/security/openness.

        I wish I could say this dogmatically, but I have some emotional need for speed. Perhaps I will grow out of it, but I doubt it.

        A RISC V development board is $1000 and a basic RISC V workstation motherboard from Raptor/Talos is $2500 and up. My entire day-to-day workstation--a Ryzen R7-1700X system tricked out with SSDs and 24GB RAM--cost less than either of these boards alone.

        I did source my small-board systems based on the criteria of "no-binary-blobs*", "must-run-unmodified-free-software*," and ended up buying the NanoPC T3 and two Olimex Olinuxino Lime 2 boards. (The Olinuxinos boot and run with no binary blob but the free driver for their GPU is only rudimentary. That was fine with me as I run them headless anyway.)

        If I won the (sweepstakes|lottery|etc) I'd ditch the Ryzen for one of those Talos boards when/if they ever come available.

        --------------------
        * Both criteria that the inexplicably popular "raspberry pi" series of devices fail to meet.

        • (Score: 2) by requerdanos on Thursday February 08 2018, @12:53AM

          by requerdanos (5997) Subscriber Badge on Thursday February 08 2018, @12:53AM (#634579) Journal

          RISC V workstation motherboard from Raptor/Talos

          I have no idea why I said that. Raptor/Talos is a Power 9, *not* a "RISC V". Got that stuck in my head thinking about the little development board, I guess.

      • (Score: 3, Insightful) by MichaelDavidCrawford on Thursday February 08 2018, @01:26AM (3 children)

        by MichaelDavidCrawford (2339) <mdcrawford@gmail.com> on Thursday February 08 2018, @01:26AM (#634601) Homepage Journal

        I am afraid I am the bearer of sad tidings:

        What the CPU vendors regard as "a lot of orders" come most commonly from the type of people who hang out on Facebook because they can't figure out email.

        Compared to them, the LUNUX neckbeards that care about security and openness are few and far-between.

        --
        My United States Social Security Number Is 518-92-8663
        • (Score: 3, Insightful) by edIII on Thursday February 08 2018, @01:43AM (2 children)

          by edIII (791) Subscriber Badge on Thursday February 08 2018, @01:43AM (#634611)

          Yes. Which is why we should all revolt.

          I'm serious. You're correct about the market, but incorrect in that the Linux/BSD/PC tech crowd (all of us) now want this openness. In the past it was largely about principles, but security over the last 24 months has taught a lot of people through fire that security is now the 1st item. Our entire house of cards rests on perilous foundations. Virtual is no longer the impenetrable wall people once thought it was, and at every level (Joe Sixpack to Sony exec), people are learning that privacy might be important.

          Some us are willing to do something about it, and pay something for it too. Purism is proving successful at finding a market for it. What will really open up the market is when more of us start refusing to service Windows 10/S, or refuse to service a proprietary board. That's taking it pretty far, but it can start slow just by replacing what the family and friends have. These days with security being what it is, I think you would find an easier time selling it.

          Perhaps we should start by figuring out what reasonable kind of fabrication techniques we could use to bake our own chips. I remember an interesting conversation about what we would need to create, from scratch, in order to build a fully working computer again. Some people were saying that specially designed 3D makers could get us a working processor. Not a very fast one, but a working one.

          • (Score: 2) by MichaelDavidCrawford on Thursday February 08 2018, @06:15AM

            by MichaelDavidCrawford (2339) <mdcrawford@gmail.com> on Thursday February 08 2018, @06:15AM (#634720) Homepage Journal

            but a lathe can't make a lathe.

            --
            My United States Social Security Number Is 518-92-8663
          • (Score: 3, Insightful) by requerdanos on Thursday February 08 2018, @01:47PM

            by requerdanos (5997) Subscriber Badge on Thursday February 08 2018, @01:47PM (#634910) Journal

            I remember an interesting conversation about what we would need to create, from scratch, in order to build a fully working computer again.

            Well over 99% of us--that includes me and probably you--wouldn't be able to make a #2 pencil without large-scale research and collaboration.

            Much less trained silicon.

            we should all revolt.

            Hear, hear. That's a movement I can get behind.

    • (Score: 0) by Anonymous Coward on Wednesday February 07 2018, @08:21PM (1 child)

      by Anonymous Coward on Wednesday February 07 2018, @08:21PM (#634547)

      I'm guessing that the problem is that your cores are starved.

      • Your software (Linux scheduler, individual programs, etc.) isn't using multiple cores optimally.

      • The hardware (cache layout, busses, storage devices) isn't supply the cores with data optimally.

      You can do a lot with 1 core at 1.4 Ghz; you can do a lot more with 8 of them.

      • (Score: 2) by requerdanos on Wednesday February 07 2018, @08:44PM

        by requerdanos (5997) Subscriber Badge on Wednesday February 07 2018, @08:44PM (#634559) Journal

        The hardware (cache layout, busses, storage devices) isn't supply the cores with data optimally.

        This is definitely a problem with this board.

    • (Score: 2) by LoRdTAW on Wednesday February 07 2018, @08:28PM (1 child)

      by LoRdTAW (3755) Subscriber Badge on Wednesday February 07 2018, @08:28PM (#634551) Journal

      Those octo-core cpu's with slow single channel memory are a joke. They shouldn't even exist.

      • (Score: 3, Interesting) by requerdanos on Wednesday February 07 2018, @08:42PM

        by requerdanos (5997) Subscriber Badge on Wednesday February 07 2018, @08:42PM (#634556) Journal

        Those octo-core cpu's with slow single channel memory are a joke.

        It's 64-bit and no slower than my olinuxinos, which are 32-bit. It didn't cost any more either.

        But the hardware configuration was made with some very, very strange decisions, none contributing to speed.

    • (Score: 2) by TheRaven on Thursday February 08 2018, @09:58AM (2 children)

      by TheRaven (270) on Thursday February 08 2018, @09:58AM (#634800) Journal
      I'd never heard of that board, but apparently it uses the Cortex A53. This is the slowest of the ARMv8 cores on the market. It is intended for two use cases: places where performance doesn't matter at all, and as the 'LITTLE' core in big.LITTLE configurations so it can run non-CPU-intensive things without waking up the fast core. You seem to be comparing a dual-issue in-order 1.4GHz chip to an out-of-order superscalar chip and deciding that it's going to be 2.4x as fast. This makes as much sense as saying that a 2.2GHz Xeon will be 2.2x as fast as a 1GHz AMD Geode (after all, they're both the same instruction set and comparing GHz numbers gives an accurate reflection of performance!). In other words, it's complete nonsense.

      In case you aren't aware of quite how much nonsense you're spouting, the Cortex A75 (the fastest chip designed by ARM, though not the fastest ARM core currently on the market) is around four times the performance per clock as the A53. The A53 isn't even the fastest of the simple in-order ARMv8 cores (the A55 is about 20% faster at the same clock speed).

      The A53 is the core that you'll find four of in the RPI3, which costs approximately nothing. It's not a high performance core.

      --
      sudo mod me up
      • (Score: 0) by Anonymous Coward on Thursday February 08 2018, @12:56PM

        by Anonymous Coward on Thursday February 08 2018, @12:56PM (#634860)

        Thanks for filling that in. I knew the A53 was slower than the A75 at the same clock speed, but I didn't know the difference was > 2x.

      • (Score: 2) by requerdanos on Thursday February 08 2018, @01:53PM

        by requerdanos (5997) Subscriber Badge on Thursday February 08 2018, @01:53PM (#634913) Journal

        you aren't aware of quite how much nonsense you're spouting

        My awareness is probably higher than you think :) Saying that this cheapie chip (it was less than $100 board and all) compares to Ampere's 8th wonder of the world is, however, probably not much more ridiculous than Ampere saying that their chip is "unsurpassed" by anything on the market. That's kind of the point, that I admittedly made poorly.

  • (Score: 4, Interesting) by Azuma Hazuki on Wednesday February 07 2018, @09:07PM

    by Azuma Hazuki (5086) Subscriber Badge on Wednesday February 07 2018, @09:07PM (#634571) Journal

    Does this thing have anything equivalent to Intel ME on it? Because if so, hard pass. If not...well, Linux runs on ARM :) Might make an interesting home desktop machine around ~2020

    --
    I am "that girl" your mother warned you about...
  • (Score: 4, Interesting) by MichaelDavidCrawford on Thursday February 08 2018, @01:23AM (3 children)

    by MichaelDavidCrawford (2339) <mdcrawford@gmail.com> on Thursday February 08 2018, @01:23AM (#634596) Homepage Journal

    The brits just wanted an inexpensive computer for schoolchildren without having to buy Z80s from the yanks.

    --
    My United States Social Security Number Is 518-92-8663
    • (Score: 2, Interesting) by Anonymous Coward on Thursday February 08 2018, @03:05AM (1 child)

      by Anonymous Coward on Thursday February 08 2018, @03:05AM (#634652)

      ARM is a disaster.

      It's an enormous blemish in the Linux repository, and it's basically useless without a pile of shite from "vendors", especially with regard to GPUs.

      Thanks, Brits!

      • (Score: 2) by requerdanos on Thursday February 08 2018, @01:59PM

        by requerdanos (5997) Subscriber Badge on Thursday February 08 2018, @01:59PM (#634917) Journal

        useless without a pile of [binary blobs] from "vendors", especially with regard to GPUs.

        While this is true especially in the case of the raspberry pi series, which require a GPU blob even to boot, not all the ARM chips are created equal.

        There are ARM chips (and onboard ARM GPUs) that don't require binary blobs to function.

        ARM as an architecture is more of a guideline than a rigid single specification. As a manufacturer, you tick the boxes you want, maybe make some stuff up, and roll your own design when making an "ARM" chip. Wikipedia calls it, rather poetically, a "family of reduced instruction set computing (RISC) architectures for computer processors, configured for various environments" which says it nicely.

    • (Score: 2) by TheRaven on Thursday February 08 2018, @10:05AM

      by TheRaven (270) on Thursday February 08 2018, @10:05AM (#634801) Journal
      Uh, no. The BBC Micro (and Acorn's other early computers) used a 6502, not a Z80. When Acorn looked for a 16-bit replacement, they didn't like many of the options. They looked at the Berkeley RISC project and saw that a team of half a dozen people using modern VLSI techniques could produce a chip that was performance competitive with ones sourced from larger companies and decided to try it. Their chip ended up being very power efficient as a result of its simplicity and so Apple became interested in using it for the Newton, but didn't want to buy chips from a direct competitor, so Apple, Acorn and VLSI Technology spun out ARM as a separate company that would sell to both Apple and Acorn (and, quite soon, lots of other companies), initially with VLSI Technology doing all of the fabrication. The chips were made in the US from the start, though they were designed in the UK. The fabrication gradually moved to Asia, but most of the design remains in the UK (though there's also a big team in Austin that does quite a lot).
      --
      sudo mod me up
(1)