Announcement

Collapse
No announcement yet.

ARM's 64-bit Juno Platform Should Be Quite Exciting

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #11
    Originally posted by mlau View Post
    I requested a qoute on a single board, ~around 5000? So you CAN get one, if you have the spare cash..
    Well, I guess, even if you are not scared after learning about the hefty price tag and are ready to pay that much money, then the corporatocrats in charge will make up some different reason to refuse selling it to you I know that at least one of the Gentoo developers tried to buy an X-GENE box for working on arm64 support and was ready to pay the requested several thousands dollars price, but this has not moved anywhere. Moreover, you will likely have to sign some binding agreements, which forbid to share the benchmark results or disclose some other information.

    So everyone has to relax and just wait for the availability of the real 64-bit ARM processors in Linux powered consumer devices, such as tablets/smartphones and low cost devboards. They may probably show up on the market in half a year or so.

    Comment


    • #12
      Originally posted by ssvb View Post
      Well, I guess, even if you are not scared after learning about the hefty price tag and are ready to pay that much money, then the corporatocrats in charge will make up some different reason to refuse selling it to you I know that at least one of the Gentoo developers tried to buy an X-GENE box for working on arm64 support and was ready to pay the requested several thousands dollars price, but this has not moved anywhere. Moreover, you will likely have to sign some binding agreements, which forbid to share the benchmark results or disclose some other information.

      So everyone has to relax and just wait for the availability of the real 64-bit ARM processors in Linux powered consumer devices, such as tablets/smartphones and low cost devboards. They may probably show up on the market in half a year or so.
      Dev hardware has a trickle effect.
      First, its to the major hardware buyers, then later on, the next layer of dev hardware from a 3rd party for consumer/hacker buyers (like IFC), then later on when its all totally obsolete, it runs down to the toy/junk sellers (rPi).

      Comment


      • #13
        Originally posted by liam View Post
        According to geekbench 3, the a7 @1.3GHz (abi compatible with arm v8a) is about 30% faster than the Core i5-2520M @ 2.50 GHz. So, clock for clock, it looks pretty good. My guess is, though, it'll be either the gen after a57, or a later rev before it gets serious. There's also the question of how fast these new archs can clock, with the best newest Intel going past 4GHz.
        I don't know what benchmarks you're looking at, but the statements above strike me as very wrong.
        A more accurate claim, IMHO, would be that (averaged over a range of code) the Apple A7 has about equivalent IPC to an i3.

        There are multiple ways to look at this. On the one hand, that's a whole lot slower than an i3 clocked twice as fast, let alone an i7 clocked at 4GHz. On the other hand, the A7 is a whole lot cheaper and smaller, and uses less power (at the same frequency, and vastly less than a 4GHz CPU).

        Basically the only honest thing to say is that the A7 is a remarkably impressive effort for only the second Apple core.
        Extrapolating beyond that (which is what people are essentially doing when they perform say an i7@4GHz comparison) heads into territory that NO-ONE knows anything about, so you're simply guessing. Yes, Intel can drive their CPU at 4GHz and Apple can't, so advantage Intel. But maybe Apple could? Who knows? They just have no need to do so.
        Maybe Intel's advantage is all process, and Apple (or at least their manufacturing partners) will get to Intel's level of process soon enough, given their higher volumes?
        Maybe Intel's advantage is circuits, and Apple does not have that sort of expertise, so they'll be at around half to a third of Intel's max frequency for quite a few years?
        Maybe Apple have a deliberate plan in mind, a specific ordering by which they will add things to their SoC (2014 means higher frequency, HW TM, better uncore; 2015 means revised micro-architecture that's 30% higher IPC than today and quad-core; 2016 means kilo-instruction-processing techniques and performance equivalent to Intel at half the frequency)? They're following the plan that gets them to where they want to be with the most appropriate CPUs at each point along the way.

        It's fun to speculate and all, but let's be realistic here. None of us have a clue what Apple's end goal is: (Replace the CPUs in all Macs with an Apple SoC? Replace the CPUs in Apple's server centers with an Apple SoC? Sell an Apple CPU to the world and compete with Intel?) let alone the capabilities of their manufacturing partners. Which means that extrapolations (implied or otherwise) beyond the actual facts --- the actual frequencies Apple ships, for the actual target products Apple sells --- tell us nothing except that Apple targets phones (where Intel is largely unsuccessful) and Intel targets servers (where Apple has zero presence today, and no claim of a presence tomorrow).

        Comment


        • #14
          Originally posted by ssvb View Post
          Well, I guess, even if you are not scared after learning about the hefty price tag and are ready to pay that much money, then the corporatocrats in charge will make up some different reason to refuse selling it to you I know that at least one of the Gentoo developers tried to buy an X-GENE box for working on arm64 support and was ready to pay the requested several thousands dollars price, but this has not moved anywhere. Moreover, you will likely have to sign some binding agreements, which forbid to share the benchmark results or disclose some other information.

          So everyone has to relax and just wait for the availability of the real 64-bit ARM processors in Linux powered consumer devices, such as tablets/smartphones and low cost devboards. They may probably show up on the market in half a year or so.
          The quote was from an official local ARM dealer. For the 5k€ I expect decent documentation about all parts on the board (main cpu register description, board schematics, datasheets for all other chips on it, The price is in line with what we paid for development boards and vendor support 10 years ago).

          Sometimes I miss my old job: I always got halfway decent documention from vendors if I asked, even from nvidia (their initial
          tegra 2 documentation they sent me didn't even have half the registers and bits of half the blocks documented. Our impression was that they thought everyone was
          going to copy their reference design and all they needed to provide was a kernel binary. The recent Jetson-TK1 documentation, or lack thereof, makes me think they
          haven't changed their thinking in the last 6 years).

          Comment


          • #15
            Originally posted by name99 View Post
            I don't know what benchmarks you're looking at, but the statements above strike me as very wrong.
            A more accurate claim, IMHO, would be that (averaged over a range of code) the Apple A7 has about equivalent IPC to an i3.

            There are multiple ways to look at this. On the one hand, that's a whole lot slower than an i3 clocked twice as fast, let alone an i7 clocked at 4GHz. On the other hand, the A7 is a whole lot cheaper and smaller, and uses less power (at the same frequency, and vastly less than a 4GHz CPU).

            Basically the only honest thing to say is that the A7 is a remarkably impressive effort for only the second Apple core.
            Extrapolating beyond that (which is what people are essentially doing when they perform say an i7@4GHz comparison) heads into territory that NO-ONE knows anything about, so you're simply guessing. Yes, Intel can drive their CPU at 4GHz and Apple can't, so advantage Intel. But maybe Apple could? Who knows? They just have no need to do so.
            Maybe Intel's advantage is all process, and Apple (or at least their manufacturing partners) will get to Intel's level of process soon enough, given their higher volumes?
            Maybe Intel's advantage is circuits, and Apple does not have that sort of expertise, so they'll be at around half to a third of Intel's max frequency for quite a few years?
            Maybe Apple have a deliberate plan in mind, a specific ordering by which they will add things to their SoC (2014 means higher frequency, HW TM, better uncore; 2015 means revised micro-architecture that's 30% higher IPC than today and quad-core; 2016 means kilo-instruction-processing techniques and performance equivalent to Intel at half the frequency)? They're following the plan that gets them to where they want to be with the most appropriate CPUs at each point along the way.

            It's fun to speculate and all, but let's be realistic here. None of us have a clue what Apple's end goal is: (Replace the CPUs in all Macs with an Apple SoC? Replace the CPUs in Apple's server centers with an Apple SoC? Sell an Apple CPU to the world and compete with Intel?) let alone the capabilities of their manufacturing partners. Which means that extrapolations (implied or otherwise) beyond the actual facts --- the actual frequencies Apple ships, for the actual target products Apple sells --- tell us nothing except that Apple targets phones (where Intel is largely unsuccessful) and Intel targets servers (where Apple has zero presence today, and no claim of a presence tomorrow).

            I said exactly which benchmark I was looking at. Did you even Google it?
            All I claimed is what the benchmarks relayed. No more.
            On single threaded perf, the [email protected] is about 30% faster than the [email protected]. That number is a composite that looks at FP, integer and memory tests.
            So, before you accuse someone of guessing perhaps your should at least inform yourself of what they are speaking about.
            Look at the site and get back to me with particular complaints.
            Lastly,I don't give five shits about apple, I'm only using then because they've the only v8a core that's readily available, do, again, next be cautious with your assumptions.

            Comment


            • #16
              Originally posted by liam View Post
              I said exactly which benchmark I was looking at. Did you even Google it?
              All I claimed is what the benchmarks relayed. No more.
              On single threaded perf, the [email protected] is about 30% faster than the [email protected]. That number is a composite that looks at FP, integer and memory tests.
              So, before you accuse someone of guessing perhaps your should at least inform yourself of what they are speaking about.
              Look at the site and get back to me with particular complaints.
              Lastly,I don't give five shits about apple, I'm only using then because they've the only v8a core that's readily available, do, again, next be cautious with your assumptions.
              I have no idea why you are so angry. Or for that matter where you got your numbers from.

              Let's LOOK at the numbers:
              iPad Air at 1.4GHz. Geekbench single core rating is 1371:


              To quote from Geekbench: "Geekbench 3 scores are calibrated against a baseline score of 2500 (which is the score of an Intel Core i5-2520M @ 2.50 GHz)"
              Scaling that 2500 to 1.4GHz would give a rating of 1400. Like I said --- equal IPC.

              Don't agree with that? OK, let's try a different Intel chip.
              Benchmark results for an iMac (27-inch Late 2013) with an Intel Core i5-4570 processor.

              This gives us a single core score (for a Haswell i5 iMac, 32-bit mode, 3.2 GHz) of 3304. Naively scaling that to 1.4GHz we get 1445. Same sort of ballpark. 64-bit mode would increase the value by a few percent, but same basic point --- a Cyclone core is about the same IPC as a an i3, and ever so slightly behind an i5 and i7 (which have larger L3 caches, probably a slightly better memory controller, and a bunch of features that help with larger benchmarks but not so much with Geekbench).

              Comment


              • #17
                Originally posted by name99 View Post
                I have no idea why you are so angry. Or for that matter where you got your numbers from.

                Let's LOOK at the numbers:
                iPad Air at 1.4GHz. Geekbench single core rating is 1371:


                To quote from Geekbench: "Geekbench 3 scores are calibrated against a baseline score of 2500 (which is the score of an Intel Core i5-2520M @ 2.50 GHz)"
                Scaling that 2500 to 1.4GHz would give a rating of 1400. Like I said --- equal IPC.

                Don't agree with that? OK, let's try a different Intel chip.
                Benchmark results for an iMac (27-inch Late 2013) with an Intel Core i5-4570 processor.

                This gives us a single core score (for a Haswell i5 iMac, 32-bit mode, 3.2 GHz) of 3304. Naively scaling that to 1.4GHz we get 1445. Same sort of ballpark. 64-bit mode would increase the value by a few percent, but same basic point --- a Cyclone core is about the same IPC as a an i3, and ever so slightly behind an i5 and i7 (which have larger L3 caches, probably a slightly better memory controller, and a bunch of features that help with larger benchmarks but not so much with Geekbench).
                You're absolutely right. I misread the baseline as being at 1000, thus my 30%, which certainly explains why you were asking which benchmark.
                I was angry for two reasons: 1. it didn't seem as if you'd looked at the benchmark site (since you were asking which benchmark), 2. got seemed to be focusing on apple when the fact that the a7 is an apple chip was entirely incidental to me. I mentioned the freq scaling only to bring up the point that even if the a7 was clock-for-clock as fast as the i5, due to it's arch it may not be feasible to scale it that high on current nodes (and without lots of cooling).
                I apologize for being short with you.

                Comment

                Working...
                X