Etc.

Happy Sunday! I’m packing to head to ARM TechCon this week and looking forward to learning some new things. The folks at ARM have been very good about engaging with the computing enthusiast community this year, as it’s become increasingly clear that their licensees will help drive the future of personal computing. They’ve opened up to us with details about their (often quite impressive) CPU and GPU architectures and other technologies, and they’re willing to share more.

With that in mind, they’ve asked me to ping you all about what you’d like to know about ARM, its technology, and its ecosystem. If you have questions about general areas or specific things, post them in the comments below, and I’ll see about getting some answers this week. I already have some questions queued up in my head about the 64-bit ARMv8 ISA and the specific bits of it that may have helped Apple achieve those very nice IPC increases in the A7 SoC. I also think big.LITTLE is an intriguing architectural choice. What else are you interested in?

Comments closed
    • the
    • 6 years ago

    What plans does ARM have for symmetric multithreading (SMT, aka Hyperthreading by Intel’s marketing). Previously ARM stated that it was unwise due to the power consumption overhead and additional cores provided a better performance per watt ratio despite the additional die space. However, the company has warmed up to the idea. Will we see SMT in ARM developed cores and if so, when? How does the performance per watt gains of SMT look for various ARM implementations, say the A7 vs. A57? How would SMT fit into ARM’s big.LITTLE plans? For example would an n-way SMT enabled big core support n little cores in an in-kernel switching (IKS) scenario?

    ARM has been releasing a cores that cover a wide gamut of performance normalized for clock speed. The A7, A9, A12 and A15 are very similar at the instruction set level. The A7 and A15 are identical to support big.LITTLE. Will we be seeing more diversity in the big.LITTLE possibilities throughout a generation of cores? IE would a hypothetical A51 work in a big.LITTLE configuration with the announced A53? Would a future A59 work with the announced A53? Long term, would we have a hypothetical A61, A63, A65 and A67 all launching simultaneously to cover the entire performance per watt and performance per die space spectrum?

    One of the oddities of big.LITTLE thus far is that each core design is in its own cluster with its own L2 cache. This appears to be a waste of both power and die space due to duplication of data between the two caches. Why can’t both the big core, the little core and a slice of L2 cache be in their own cluster to reduce transistor count? Would putting the big and little cores closer lower the latencies involved for transitioning to the big core under an in-kernel switching scenario?

    Under the current multiple cluster scenario for big.LITTLE, can the L2 cache be different sizes? IE can the big cores have 2 MB L2 cache shared in the cluster versus 512KB shared for the little cluster? What would dictate what gets transferred from the big cache’s domain to the little core domain?

    Process technologies are allowing designs to embed large amount of memory on-die either as DRAM or SRAM. For the embedded markets ARM has geared many of it designs for, the capacity obtainable on state of the art processes would be enough to forgo external memory entirely if willing to pay for a large die. Has ARM considered the possibility of adding an eDRAM controller to its IP portfolio? If so, was it geared for low power consumption or a higher performance, ultra wide (greater than 512 bit) interface? Would ARM support this in a fixed capacity or let SoC designers configure the amount of eDRAM? By effectively removing the memory latency and bandwidth bottleneck, what performance gains would be gained in the ARM ecosystem? Say would an A5 perform like an A7 at similar clocks? A7 to A9?

      • UberGerbil
      • 6 years ago

      WRT SMT, see [url=http://www.broadcom.com/press/release.php?id=s797235<]this Broadcom press release[/url<] -- 4 threads per core, aimed at the server market (though the PR is pretty short on technical meat). Probably won't see it in the wild until at least 2015, though.

        • the
        • 6 years ago

        Nice, that announcement slipped past my radar. Considering it is referencing 16 nm production, chances are that it won’t appear in products until late 2016.

    • heinsj24
    • 6 years ago

    I’d be interested in hearing about HSA and ARM. We are always hearing about AMD and HSA, but never ARM, despite 4 of the 7 founding members producing or designing ARM processors.

    • Igor_Kavinski
    • 6 years ago

    I want to know if ARM has any long term plans to add AI to their CPU cores. I’m sure that they don’t really intend to keep churning out dumb processors forever. They have access to great minds who helped forge the end of Intel’s dominance in personal computing by heralding a new era of smart personal devices. I would love to get some idea about the disruptive computing concepts that they or their partners are developing right now that will be relevant 10 years from now.

      • Scrotos
      • 6 years ago

      Your ideas are intriguing to me and I wish to subscribe to your newsletter.

    • Steele
    • 6 years ago

    Do they have any comment on the up-and-coming LEG technology?

    /sad trombone

      • Pwnstar
      • 6 years ago

      Rename MIPS to LEG! Brilliant!

    • vvas
    • 6 years ago

    Not really a question, but the more information you can get about the imminent A12, and also about A53/A57 coming after it, the better!

    • ColdMist
    • 6 years ago

    Do you you get specific design improvements back from Apple/Qualcomm/nVidia/AMD/etc? If so, do you integrate those into your designs?

    Ie, is it a 2-way street, or is it purely a 1-way street, as far as the ever-improving design goes?

    • brucethemoose
    • 6 years ago

    Apple and Qualcomm are arguably THE two major high end ARM players atm, almost everyone else already dropped out (TI, ST-Ericsson, Broadcom, possibly Samsung, etc.) while a select few are fighting to stay relevant and/or pursuing niches (Nvidia, Calxeda, AMD, others). Yet Apple’s and Qualcomm’s designs are diverging farther from ARM’s with each generation… I doubt they would care to comment on that, but I’d like to know what they think.

      • Pwnstar
      • 6 years ago

      I don’t understand. Is Samsung not “high end” anymore?

      [quote<]possibly Samsung[/quote<]

        • adisor19
        • 6 years ago

        Samsung doesn’t design their own CPU cores like Qualcom and Apple does. They take ready to implement cores from ARM and they complement them with various other IP blocks (GPU, Audio etc.) to form their SoCs.

        Adi

          • Pwnstar
          • 6 years ago

          That is one definition of “high end”, I guess. What definition does brucethemoose intend?

            • torquer
            • 6 years ago

            In the context of the article its plainly obvious he’s talking about those companies who are actively pursuing high performance custom ARM cores. There is an obvious and measurable delta between Snapdragon and Exynos on a performance basis.

            • Pwnstar
            • 6 years ago

            You imply Samsung = MediaTek. I don’t think that is true.

            • brucethemoose
            • 6 years ago

            Actually, I meant that Samsung appears to have abandoned Exynos, hence “possibly”. Despite having an in house SOC, they choose to use Qualcomm in the majority of their smartphones and other products. I’m not sure if that’s right, but to me, choosing the competition over thier own design suggests that designing SOCs is no longer a priority/

    • Applecrusher
    • 6 years ago

    If you could go back 5/10 years and change 1 decision that was made, what would it be and why?

      • brucethemoose
      • 6 years ago

      “We should’ve bought more patents!”

    • codedivine
    • 6 years ago

    Would love to hear more about their plans on the graphics side as well. Particularly:

    a) What are their plans for HSA? Do they plan to introduce HSA compliant cores anytime soon?
    b) Do they plan on supporting OpenCL on Android? Google seems to not be too interested in OpenCL. What about ARM?
    c) Do they think GPGPU and heterogeneous computing will play a role in future ARM based servers?

    • jdaven
    • 6 years ago

    ARM is the Intel of the Post PC era.

      • NeelyCam
      • 6 years ago

      Intel will be the “Intel” of the Post PC era

        • DeadOfKnight
        • 6 years ago

        Yeah. I don’t see them going up in flames regardless of the fate of PCs.

      • Krogoth
      • 6 years ago

      You couldn’t be any further from the mark.

      Intel is already catching up and it will not be much longer until ARM becomes besieged like AMD.

      Intel isn’t called Chipzilla and 800-pound gorilla for nothing.

        • jdaven
        • 6 years ago

        Your reasoning would also extend to MS but they have been trying for two years now to get a foothold in the smartphone/tablet market with no success. As a computer nerd, I know the safe bet in your mind is Intel or bust. But no one cares about Intel in the smartphone/tablet world. Intel has been trying for the last two years as well and their marketshare is even less than what MS has achieved (unless you can show me a link that says otherwise).

          • Krogoth
          • 6 years ago

          Intel is catching up. They have a massive advantage that ARM lacks, semiconductor production technology. Intel is the leader in the semiconductor business. They are taping out real products with 14nm node, while the rest of the industry is still at 22-32nm.

          Their CPU architects and designers are only beginning to get the hang of designing ultra-low voltage chips. ULV Haswell chips are nearly almost as efficient as their ARM counterparts. I would imagine that ULV Broadwell will bridge the gap. Intel has the production capacity and engineering talent to make OEM for embedded systems and ULV to desire their platforms. I’ll not be surprised that Apple will completely switch over Intel stuff in their next generation iGadgets.

          The only thing that ARM did is made Intel give a crap about doing ULV designs rather being stuck as some side project on the backburner.

      • chuckula
      • 6 years ago

      So if ARM is the Intel of the Post-PC era, and you don’t like Intel, then you can’t like ARM and you have to root for the underdog.

      But… the underdog is Intel….

      So logically… because you hate Intel you have to be an Intel fanboy!
      AND YOU WEIGH AS MUCH AS A DUCK!

    • dragmor
    • 6 years ago

    1) Design Power / Performance versus OEMs Shipped values
    What is ARM’s opinion of the OEMs shipping chips so far out of the the main distribution of the power / performance bell curve i.e. 1.9ghz vs the recommended 1.3ghz? Is ARM worried about losing the low power and stability perception due to the OEM overclocking?

    2) What metrics are you targeting for future chips?

    3) In hind sight how different were your estimated targets for A53 / A57 to what you would have aimed for if you knew what you know now.

    4) Is big.LITTLE really worth the effort compared to component gating? The context switch penalty seems to be quite large especially since the switch is done in software and not built into the hardware.

    • Flatland_Spider
    • 6 years ago

    What plans are in place to get ARM boards into the mainstream, and what plans are in place for ARM boards to interact with the greater PC ecosystem? There is the Raspberry Pi, Beagleboard, ODROID, but none of them are really equivalent to a random ATX/mATX/mini ITX motherboard on Newegg. For instance, the Raspberry Pi hangs the Ethernet controller off of the USB port rather then a PCIe bus, none of those boards have SATA/mSATA ports, and there aren’t any provisions for addin cards like video cards, WNIC/NIC, RAID card, or anything else that plugs into PCIe slots.

    When is the bootstrapping issue going to be fixed, and how are they going to do it? Right now Red Hat is pushing UEFI as the solution in the server space, and I haven’t heard anything about a similar push in consumer electronics. Coreboot is GPLed. is that an option?

    Why isn’t ARM using a FOSS license, preferably BSD or MIT, for their Mali graphics driver? I understand there is a lot of proprietary stuff that goes into graphics, but a FOSS driver with basic support would be nice.

    What are some good resources for getting into ARM development and/or getting a Linux distro running from scratch? I like this sort of thing, and making it easy to get people programming on the platform is necessary.

      • willmore
      • 6 years ago

      I have been asking for a normal PC sized MB with an ARM chip on it (and normal PC I/O) for years. Someone make this!

      My last router build went to a Mini-ITX E-350 board. It wasn’t my first choice, but it was something I could buy. Come on, people, make some boards!

    • jjj
    • 6 years ago

    Why aren’t they even trying to go after the desktop and desktop gaming ,someone should get things rolling, hardware and software.
    How will they avoid consolidation in the phone/tablet/wearables market in the long run, short term it’s clear how they fight it.
    Should we expect them to go for more and more types of computing units ,including baseband?
    Do they cooperate in any way with any security agencies?
    If one day we’ll be able to print chips at home ,how do they plan to handle that market? – that’s a bit too far ahead.
    Do they have any comment on Qualcomm Zeroth?

    • Beelzebubba9
    • 6 years ago

    How does ARM intend to maintain a Power:Performance lead* now that Intel has thrown their weight behind low power SoCs and looks to retain it’s 12-36 month lead on advanced process technologies?

    *(not that any ARM designed CPU has a real power:perf advantage over Bay Trail currently)

      • jdaven
      • 6 years ago

      The same way Intel managed to retain 80% + marketshare when AMD was blowing the Pentium 4 out of the water with the Athlon 64…ARM owns the market.

    • sschaem
    • 6 years ago

    1. If ARM is going to enter the x86 turf at some point, is there any good metrics to measure progress?

    Example, is there any version of POVray running on ARM ? and result published.

      • chuckula
      • 6 years ago

      While phoronix’s testing methodology is far from perfect and shouldn’t be taken as the final word on performance, it does have quite a few benchmarks that compare ARM to x86 parts (and different ARM parts to each other) in various Linux benchmarks.

      Here’s a recent link comparing a quad-core A15 to older-generation Atoms and some other ARM chips: [url<]http://www.phoronix.com/scan.php?page=article&item=arm_odroidxu_octa&num=1[/url<]

      • Pwnstar
      • 6 years ago

      Do you not like the JavaScript comparisons?

        • UberGerbil
        • 6 years ago

        There are so many layers to that it’s hard to know what you’re comparing, unless you’re working from exactly the same codebase and everything in the toolchain on both platforms has equivalent maturity and optimizations.

    • chuckula
    • 6 years ago

    1. Since the vast majority of applications that use ARM’s Mali GPU cores are using some form of Linux (Vanilla or an Android variant), when can we expect to see open-source GPU drivers that are contributed to the mainline kernel in a similar manner to how Intel’s GPU architecture (including the Baytrail GPU) has open-source kernel drivers?

    2. Did Apple use a directly ARM-derived version of the A53 or A57 in the A7 (Massive use of As there!), or is the core radically different from those designs in a similar manner to how the A6 or Snapdragon cores diverge from the stock ARM implementations?

    3. Qualcomm is ARM’s biggest licensee.. but in some ways is also the biggest competitor against other ARM licensees. Given that Qualcomm (and also Apple) have expressly decided against adopting big.LITTLE and that Samsung has had issues with its own big.LITTLE implementations, what is ARM doing to make this solution more attractive to other licensees?

    3.5 Related to point 3: Does ARM make more money for a big.LITTLE chip compared to a chip that is just big or just LITTLE? Could pricing… in addition to the added complexity, be detracting from the success of this approach?

    4. Given that the Cortex A15 is showing its age in the performance-to-battery life ratios* what does ARM have in store given that ARM licensees are not getting access to the most cutting-edge lithographic processes? We have already seen the first roadmaps of 64-bit ARM server chips from AMD where 8-core A57 parts from late 2014 are projected to have a 50% higher TDP than 8-core Atom server parts that are commercially available today. ARM has a very long track record of having solid low-power designs, but what assurances do we have that ARM can keep those power saving features while trying to scale to higher performance envelopes?

    * See Arstechnica’s two interesting reviews of Chromebooks where Haswell is, even given a larger battery, at least at parity with the Samsung Exynos A15 while delivering roughly twice the performance in a lower-price chromebook:

    [url<]http://arstechnica.com/gadgets/2013/10/review-hps-279-chromebook-11-raises-an-admittedly-low-bar/[/url<] [url<]http://arstechnica.com/gadgets/2013/10/review-acer-and-haswell-give-chrome-os-the-battery-life-it-deserves/[/url<]

      • zdw
      • 6 years ago

      re: #4, wouldn’t the A12 be a response to this?

        • Beelzebubba9
        • 6 years ago

        The A12 is a slower, cheaper core design to sell alongside the A15 in price sensitive designs. I don’t see how that will change the price:perf landscape against Avoton.

      • brucethemoose
      • 6 years ago

      Ya, I wonder how ARM can even respond to broadwell when it comes around.

        • NeelyCam
        • 6 years ago

        Better, in terms of battery life or performance. A53/A57 are more optimized for power-efficient performance, and coupled with TSMC 20nm, ARM may finally reach “good-enough” performance while maintaining power efficiency.

        The performance and power-efficiency delta between Intel and TSMC is likely to remain roughly the same (22nm Intel Trigate vs. 28nm TSMC planar, 14nm Intel Trigate vs. 20nm TSMC planar).

        And feel free to substitute “TSMC” with “Samsung”; they are probably going to hit 20nm roughly around the same time.

      • jus10
      • 6 years ago

      Yes, definitely question #1 above. Are we ever going to see drivers for the Mali? I have a Mali T604 in my Chromebook and the undocumented binary only incomplete releases to date aren’t really adequate. I’m not looking to to play Quake, but I’d like basic acceleration and specifically video acceleration.

      I’d just like to know if I should hold out hope or let it go!

      • the
      • 6 years ago

      2. It has been confirmed that the Apple A7 is using a design developed in house by Apple themselves. One of the reasons Apple went this route is time-to-market, which they beat ARM properly with the first commercial 64 bit chip.

        • willmore
        • 6 years ago

        Cite sources.

          • windwalker
          • 6 years ago

          Apple, you nitwit.
          Oh, I forgot. They are evil so they always lie.

          • Beelzebubba9
          • 6 years ago

          I’m not sure anyone knows for sure – but I’d probably heap my guess in with Anand and say Cyclone is an evolution of Swift.

          Hardly a sound source, I know. 🙂

          • the
          • 6 years ago

          [url<]http://anandtech.com/show/7335/the-iphone-5s-review/3[/url<]

        • Pwnstar
        • 6 years ago

        But ARM doesn’t make chips, so how exactly could they be beaten? The A7 uses ARM’s ARMv8 anyway, so I don’t understand your point.

      • windwalker
      • 6 years ago

      Any other more loaded questions and inquiries about information that the company obviously must keep secret?
      You can do better than that.

        • chuckula
        • 6 years ago

        Nice personal attack there.

        Tell you what, since you are so much smarter than everyone else on these forums, why don’t you answer just one of those questions? Come on, you are at least twice as smart as anyone at ARM, so it should be easy! Go ahead, I’m waiting.

Pin It on Pinterest

Share This