Can someone explain why ARM is so efficient compared to x86?

Can someone explain why ARM is so efficient compared to x86?

It's All Fucked Shirt $22.14

UFOs Are A Psyop Shirt $21.68

It's All Fucked Shirt $22.14

  1. 1 month ago
    Anonymous

    It has estrogen cores like intel started adding to their x86 cpus. Except arm was built around hybrid cores from the start.

  2. 1 month ago
    Anonymous

    x86 is a rhino aged boomer chipset that contains instructions on top of instructions on top of instructions. ARM has a smaller set of instructions that can get the same tasks done more efficiently. It's simply better.

    • 1 month ago
      Anonymous

      why are all desktop parts still x86 though

      • 1 month ago
        Anonymous

        Because until recently it's been a duopoly with nobody too interested to improve things beyond the bare minimum.

        • 1 month ago
          Anonymous

          As far as I am concerned, it still is a duopoly. We got ARM and x86. RISC is ok, but I don't see anyone using it outside of a few embedded deviced yet. Perhaps one day we will get that third option.

      • 1 month ago
        Anonymous

        it has a smaller instruction set, making it easier to implement in silicon and less complex. This makes it faster at lower power. However this also makes it unable to scale as high as x86, because it has to break down big tasks into more small tasks while x86 has esoteric used-only-once-per-day instructions that can handle some extremely specific vector manipulation tasks in a single cycle. It's all a trade off.

        legacy compatibility is a big thing.
        Also x86 is still king when it comes to performance per physical space, which is what matters for things like server rooms and supercomputers.

        • 1 month ago
          Anonymous

          >which is what matters for things like server rooms and supercomputers.
          Why are they moving to ARM too then?

        • 1 month ago
          Anonymous

          >x86 has esoteric used-only-once-per-day instructions that can handle some extremely specific vector manipulation tasks in a single cycle.
          I think you're confusing two different kinds of instructions. The funny CISC used-only-once-per-day instructions (eg. bt mem32) absolutely do not take "a single cycle." Most of them aren't used anymore because they're slower than open coding what they do with simpler instructions. The "vector manipulation" instructions are different. They're from much more modern extensions that are basically the same as the equivalent ARM extensions in power.

        • 1 month ago
          Anonymous

          literally all supercomputers are ARM lol

          • 1 month ago
            Anonymous

            Most of the Nov 2023 TOP500 supercomputers are x86. The efficiency gap between Fugaku (ARM) and Frontier isn't too bad neither, and both use TSMC 7nm so it should more or less be a level playing field minus things like Frontier having GPUs while Fugaku doesn't have any kind of GPU that I can tell.

        • 1 month ago
          Anonymous

          >However this also makes it unable to scale as high as x86, because it has to break down big tasks into more small tasks while x86 has esoteric used-only-once-per-day instructions that can handle some extremely specific vector manipulation tasks in a single cycle. It's all a trade off.
          Modern x86-64 code is fairly simple and only occasionally uses memory operands, mostly sources. ARM uses many instructions that use more that 3 registers and perform multiple operations at once.

      • 1 month ago
        Anonymous

        Because desktops are plugged into dedicated power supplies and performance matters more than energy efficiency.

        • 1 month ago
          Anonymous

          efficiency is performance

          • 1 month ago
            Anonymous

            No it's a ratio of input/output. If the input is effectively limitless, only the output matters.
            >inb4 BUT I CAN BUILD A CLUSTER OF ARM CPUS WITH ALL THE EFFICIENCY
            Ok go ahead I'll just use the single socket x86

          • 1 month ago
            Anonymous

            the input is certainly limited though, in heat generation and cost. you can't just generate images in Stable Diffusion on a 4090/14900K system for an hour during the summer time

          • 1 month ago
            Anonymous

            >efficiency is performance
            only a delusional itoddler baboon would seriously believe this. you have to believe it when you buy into horseshit products and a corporation that thinks people buy machines based on power consumption instead of raw performance. this is exactly why apple has been losing market share for years and why m4 is on PAGE 2 of geekbench's single core performance ratings.

            x86 has more instructions, but I think you underestimate the amount of bloat that has found its way into a modern ARM cpu. ARM is bloating out just like x86 did before it, x86 just has a headstart.

            there's a lot of bloat but it's not really the fault of the CPU architecture. it's manufacturers creating SoCs that are bloating the living shit out of it. and remarkably they still need a co-processor/dsp/"accelerated functions" to actually function well at all.

    • 1 month ago
      Anonymous

      That's only part of it. Both Intel and AMD focus primarily on performance when designing their chips. Lowering clock speeds and power targets can only ever partially offset that.

      • 1 month ago
        Anonymous

        >Both Intel and AMD focus primarily on performance when designing their chips.
        M1 performance is not lower than the slowest Intel chip, while it absolutely kills it at power use.

        • 1 month ago
          Anonymous

          The slowest intel chip comes from the same production run as the fastest chip. It's not designed around anything, it's production refuse that some people are willing to pay for.

          • 1 month ago
            Anonymous

            How about this: the fastest Intel CPU is twice as fast as the M1, but needs 10 time more energy.
            Imagine if they doubled the cores of the M1, it could probably be as fast as Intel while still using less power.
            Power efficiently may not be important for your gaming desktop. But the consumer desktop market is tiny. Power efficiency is important for laptop, phones, and... servers! A data center would love to more than halve its energy bill, which includes cost both for powering the servers and the cooling infrastructure.

          • 1 month ago
            Anonymous

            >A data center would love to more than halve its energy bill, which includes cost both for powering the servers and the cooling infrastructure.
            The main thing limiting how tight we can pack computers in a datacenter is cooling, i.e., literally keeping the place from cooking itself to death.

          • 1 month ago
            Anonymous

            You can't improve single core performance by increasing core count. A single threaded web server can only be made to run faster by speeding up the core, and that's where intel is king.

          • 1 month ago
            Anonymous

            Video encoding isn't single-threaded. High performance web servers aren't single threaded. Video encoding doesn't scale as well with higher core counts, but if you do segmented encoding (with shitbrake or shitshitpeg can't do automatically) you could get a lot more out of it. Or you know, just get two M1 servers instead of a single Intel one.

    • 1 month ago
      Anonymous

      x86 has more instructions, but I think you underestimate the amount of bloat that has found its way into a modern ARM cpu. ARM is bloating out just like x86 did before it, x86 just has a headstart.

    • 1 month ago
      Anonymous

      KYS
      CISC > RISC by definition

  3. 1 month ago
    Anonymous

    Shame the handbrake energy use doesn't have AMD. Would be interesting to see how they do compared to 12700 which seems decent enough at 65w.

  4. 1 month ago
    Anonymous

    >muh watts
    Low consumption comes with shit performance

    • 1 month ago
      Anonymous

      That's where you simply add more cores. Besides, 25% performance at 50% power drain just isn't worth it.

      • 1 month ago
        Anonymous

        tech illiterate take

  5. 1 month ago
    Anonymous

    >Can someone explain why ARM is so efficient compared to x86?
    ARM is the Single Sideband Redneck Radio compared to the x86 Amplitude Demodulation.

  6. 1 month ago
    Anonymous

    system efficiency =/= cpu efficiency
    Normal PC have a lot of hardware bloat and "inefficient" interfaces to improve flexibility.
    I prefer spending 3 watts more than the "holding it right" experience

    • 1 month ago
      Anonymous

      how inefficient are the other components?

      • 1 month ago
        Anonymous

        I'm not sure but I remember several examples of normal PC tweaked to achieve ultra low power consumption (idle and under load), <10 w for a desktop. In extreme cases with rarely used parts/interfaces removed.

  7. 1 month ago
    Anonymous

    >handbrake results just show Intel
    have a nice day.

    • 1 month ago
      Anonymous

      feel free to find data with AMD included. idgaf

  8. 1 month ago
    Anonymous

    no software
    a phone is efficient too

  9. 1 month ago
    Anonymous

    >5nm x64 chip more efficient than 5nm ARM chip in the middle graph
    hmm that's weird...

  10. 1 month ago
    Anonymous

    So is it really the instruction set that is causing whatever efficiency advantages Apple devices may have, or is it other factors coming into play? Remember that that Apple fully controls the hardware and software on their machines, allowing them to do all sorts of tweaks to optimize for efficiency that is not an option for other devices. Also, Apple can always afford the latest and greatest TSMC fabrication process, which of course also has a huge impact that is independent of the ISA used. For example, the M3 chips is made using TSMC's 3nm process, while that Ryzen 7840u that is only a little behind it efficiency wise in that Cinebench test is made on a 4 nm process.

    For a more Apples to Apples (lol) comparison we will have to see how well Qualcomms new Arm chips for Windows stack up against Ryzen chips being made on the same process.

    • 1 month ago
      Anonymous

      It is 5% easier decode, 5% soft/hard integration and 40% always being in the leading process and 50% simply having designed a bigger better microarchitecture

  11. 1 month ago
    Anonymous

    The ones on bottom are because these chips have die area dedicated to video encoding/decoding, which on PCs is usually on GPUs, so not really a fair comparison when the x86 CPUs are essentially bruteforcing it.

    • 1 month ago
      Anonymous

      and the Intel chips have encoding blocks too but were not used for the test
      weird

      • 1 month ago
        Anonymous

        We know why.

  12. 1 month ago
    Anonymous

    Because the x86 chips are performance focused. They simply perform better but also draw more power. It's like comparing a V12 car to a 1.0 econobox and asking why it gets lower gas mileage.

  13. 1 month ago
    Anonymous

    I just noticed those Cinebench results had a non-Apple ARM chip in them - that MS Surface with a Snapdragon chip in it. Which was dead last in multi-threaded efficiency. Goes to show that ARM is not some magic power saving panacea, and that the actual chip design and the process it is fabbed on are far more important.

    • 1 month ago
      Anonymous

      snapdragon is shit in general iirc

  14. 1 month ago
    Anonymous

    Because embedded's been the main use case for ARM for decades, that's where all the chip designers who really know about optimizing for power use work. There are lots of design choices affected by that (which combinations of transistors to use in the gates, that sort of thing; proper low-level stuff) and they've had a long time to get good at it.
    By contrast, Intel and AMD have had their bills mostly paid for by the performance-at-all-costs crowd. It's a totally different set of approaches.

  15. 1 month ago
    Anonymous

    >Can someone explain why ARM is so efficient compared to x86?
    yes, but you won't understand why it is so.
    First of all, it's AMD64, nobody runs x86 cpus anymore, even though there are some that still support IA32.
    Anyway, when you test entire systems you have to consider DDR4 vs LPDDR4, IPS vs OLED, HDDs vs SSDs, interconnects, PCI-e switches on the mainboard, interconnets on the northbridge, how general-purpose-built is one CPU vs the other and the surrounding system around it.
    Now for CPUs you have to compare cache sizes, purpose built accelerators,
    how complex is the OoO execution, how many functional units and reservation stations you have built in the cpu to compensate for bad or unoptimized software,
    how wide are your buses and how wide instructions you have to support, if I am reading the armv8 manual correctly the Q1 registers are 128bits while the 4700u has 256 wide buses for the AVX2 instuctions.
    then we go back to software, from schedulers, drivers, who writes the compilers, who targets the platforms, etc.
    Purpose-built is in general better than general-purpose, but in this case, the Apple M1/2/3/4/5/6 have an unfair advantage of operating of having morons that can't interpret what they are reading.

Your email address will not be published. Required fields are marked *