Fastest ARM chips are comparable to Intel’s slowest Atom chips

1. It will be interesting to see how the a15 fares in a few months. It’s faster, dual-core, and most-importantly Out-of-order and superscalar. Those last 2 are things ATOM doesn’t have. And ARM doesn’t need 64-bit to escape a registry glitch like x86, instead it drops to 16-bit to improve throughput. Compilers are still improving on using this trick.

   1. * Register glitch = the original x86 didn’t have enough registers which is where all performance gains come from in AMD64

2. Intel chips has compatibility quick reference guide and code name decoder, with the products including processors and desktop boards.

3. By the time Intel makes it to 14 nm, they will almost certainly be using a fraction of their manufactuing prowess to produce high-end ARM chips (to control the mobile market) while maintaining a foothold on the PC high-end.

   Companies like Intel do whatever it takes to stay ahead.  They will lie, cheat, kill, burn babies, etc.  Intel is hung up on x86 architecture only because it suits them for the moment.  As soon as their crystal ball explodes spraying rabid leeches everywhere they will adjust and flea for whatever river of blood that they can suck onto to please their shareholders.

   And the competition will only benefit the consumer.

   Nothing is stopping Intel from quickly putting together a 22nm ARM processor that sells for twice as much as Tegra 3 or 4.  For if Intel doesn’t weave something together, their bloodthirsty spiders won’t reach adulthood.

   This my friend is NOT going to happen.

4. Interessante, pena que é impossível, já fiz testes de performance e comprovei que um ARM a7 é até 5x mais rápido que um atom. Isso é igual a intel colocando um vídeo de jogo pra dizer que seu processador roda o mesmo(mas até o vídeo roda travando) como eles fizeram a menos de um mês.

   1. Não é impossível, você está confundindo ARM next gen e não o que está sendo comparado agora.

      28nm Cortex A15, com A7, vai ser até 5x geração atual.

      Medfield está provando que o átomo pode realizar a sua própria contra atuais Smart Phones.

      Enquanto ARM próxima geração vai ter que lidar com ATOM next gen no próximo ano.

5. Something that has increased in the atoms is better graphical performance in the N2xxx series.

   1. Yes, but driver support is appearing to be a problem for now.  However, Intel is going to give the ATOM a major overhaul next year with the 22nm Silvermont.

      So the ATOM will finally have Out Of Order Processing, SoC, Tri-Gate Transistors, may use a more basic version of the Ivy Bridge GMA, and a bunch of other enhancements.

6. The above post partially answers the question I had when reading the main article – which processor gets more done per watt? I’d like to see the actual numbers for that. I know ARM processors sip power in comparison. (Then throw in the surrounding chipset and ARM is so much more energy efficient). 

   For 80% of my daily tasks (reading, email, moderate web browsing, trading), the ARM in my smartphone does just as well as the i5 in my desktop, while using tons less power.

   1.  It depends on what you’re doing, ARM’s advantage is it can be designed far more simply.  So doesn’t have to support any extras that aren’t specifically needed for the device.  However, that also means less flexibility if you want to use a given device for more than it was originally intended.

      While the overhead that x86 architecture has becomes negligible as you go towards higher performance range, which ARM is starting to go into as they get ready to start competing directly with x86, like with the upcoming Windows 8 devices.

      Another factor is the silicon fabrication, back when the FAB was pretty large power leakage was negligible but when they hit 50nm it had already become nearly 40%.  So as they continue to shrink the FAB it is becoming increasingly important to control the leakage and that’s something Intel is starting to get a edge on versus ARM.

      Since ARM relies on design efficiency but has little control over the FAB, especially since many don’t make their own ARM chips but order them from the few that do manufacture the chips.

      Medfield for example shows Intel can get pretty close and if they can provide something that can compete with ARM’s next gen offerings then there is a chance that they can indeed start competiting with ARM.

7. I wonder if the H264 decoding manged to hook into the on-soc DSP in some way. There is also the neon extensions, the ARM equivalent of MMX, that accelerates SIMD style processes.

8. Encoding tests are notoriously hard to use as benchmarks.  If there is dedicated hardware in the SOC to help a particular type of encoding process, it can wildly skew a result.

   Looking at the computational data he provided, the dual core 1.2 ghz ARM Cortex A9 is anywhere between 10% to 60% slower than a single core 1.6 ghz N270.  Because the cache is so much faster on the ARM, memory intensive computational tasks are actually faster on the ARM product.  Given that the N270 is running at a 33% higher base clock frequency this means that on average the Cortex A9 has about the same IPC as the Atom…   If you consider a dual core processor the same as a single core symmetrically multithreaded one, which they clearly aren’t.  If you switch to a instructions per watt metric, ARM is demolishing that old Atom chip, as well as trouncing the rest of the processors.  So it depends on what you’re looking for I guess.  It would definitely be interesting to see how 28nm ARM A9’s compare against the latest 32nm Intel Atom chips.

   Still compared to his earlier work, this shows the impact of software optimization more than anything.  Just a couple fixes like the Hard-Float issue in Ubuntu and Linux in general, sped the ARM solution up by almost 20%.  That’s a lot of performance gain for a compiler update.  Although not as large, there are probably more gains to be had out there, and certainly more frequency on the table as maufacturing moves from 40nm down to 28nm.

   Thanks for the link, that made for an interesting read.

   1.  28nm is making the switch to Cortex A15, not A9.  It’ll definitely beat the present 32nm Medfield, with Cortex A15 providing up to 5x of today’s ARM solution performance, but Intel is switching to 22nm next year and giving the ATOM a total re-working.

      Features like Intel Tri-Gate Transistors and switching from In Order Processing to Out Of Order Processing should give them a nice performance and efficiency boost.

      While all ATOM’s will be going SoC, implementing many of the power saving methods used already by ARM (turning off anything not being used, etc.), variable TDP, faster and more memory, SSD enhancements will start becoming standard, among many other changes.

      ARM though still has it’s own performance handicaps like it’s still a 32bit only architecture.  More advance memory management is already starting to be used but true 64bit solutions are still years away. 

      ARM solutions are also by design not as flexible as x86, since ARM relies heavily on simplifying of design.  This helps gives them the advantage of being cheaper to make and more energy efficient but harder to re-purpose or add functionality that it didn’t start out with without replacing the chip with a redesigned one.

      1. Where did you get 5x performance improvement?? https://en.wikipedia.org/wiki/ARM_Cortex-A15_MPCore says 40% faster over A9

         1. It also says “all things being equal”, which they’re not going to to be when you factor all the differences.

            The TI OMAP 4460 is a up to 1.5GHz dual core based on Cortex A9 for the base comparison.

            While the Tegra 3 is more a intermittent step between present gen and next gen because it uses features like a low power 5th core to boost efficiency and is one of the first mainstream quad cores.  It is however still based on Cortex A9, similar clock speed range with a max of 1.4GHz, and it’s a 40nm FAB.

            The upcoming 28nm will add the performance boost of the more capable Cortex A15, along with using the more power efficiency A7, but you’ll be seeing quad core and even higher numbers of multi-cores becoming more the norm, especially for the server market, and the increase in efficiency combined with the FAB reduction to 28nm means we’ll also be seeing much higher max clock speeds of well into the 2GHz range.

            So add number of cores, higher clock speed, and more powerful processors and you get more than just a base 40% improvement. 

            The boost in power efficiency is also important as that means they can put more energy into performance without putting a hit on expected run time.  The A7 being of note there as it’s much more energy efficient than the A8 previously used in low power applications.  So multi-cores and higher clock speeds become more standard for even Smart Phones.

            Add improvements in technology, like faster memory, larger bandwidth, etc. Also adds more of a boost to overall performance.

            Finally, mind that my statement was for “UP” to 5x, as like now there will be a range of low offerings up to high end offerings.  So I wasn’t saying all offerings will be that much better but they can be and one of the goals of this next gen of ARM chips is to find usages in the traditional PC markets.

            Already ARM is making a impact on the server market and with Windows 8 for ARM they stand to start influencing the laptop market.

            Though keep in mind, even if they exceed the performance of the ATOM, it’s still the low end compared to the Intel’s Core i-Series that starts at over triple that level of performance. While both the ATOM and Intel’s higher end chips are steadily improving each year, with the ATOM getting a major overhaul next year when it hits 22nm FAB, and will achieve another FAB shrink to 14nm in 2014.  So there is still a chance that the ATOM can regain its performance lead. 

            Intel is just a bit late about going about it as previously they had the ATOM on a slow 5 year product cycle but with 22nm Silvermont they officially put the ATOM on the same 2 year cycle as their higher end offerings.

            Even Medfield is still based on pretty much the same ATOM architecture as when the ATOM was first released.  So ARM is basically just catching up to old low end Intel technology for performance.  Though they still got a good lead on Intel for lower cost and energy efficiency and that’s somethign even the next gen Intel chips may not overcome, but if they regain the performance lead then they just need to get close enough and while old tech the Medfield shows they can at least get close to ARM power efficiency.

            1. Ok, so its 40% for architectural change, 25% for increased clock speed and 50% (at best!) for increased number of cores. That is still way below 5x improvement.

               Finally, I like your argument with the “UP” thing. Though in that sense next gen Atoms will be UP to 10x faster 🙂

            2.  No, the clock speed increase is up to triple of what is available now. Mind many phones are still 1GHz.  Also, you’ll see more than a 50% boost going from dual core to 8 cores.  50% is only when number of cores are only doubled but it’s even more when quadrupled.

               Though those may only be for ARM laptops/tablets and servers and Smart Phones may only go up to quad cores and maybe only double the clock speed.

               While another factor not listed before is Android 4.0 on up adds better support for multiple cores for a better boost when adding more cores.

               Many apps still don’t take advantage of more than one core, but this will start changing in time for the next gen systems.  Along with a increased use of hardware acceleration.

               Though you may be right about the next gen ATOMs, Intel is really pulling all the stops on improving the architecture with the 22nm Silvermont.

            3.  You are comparing multi cores of the Tegra platform to the Atom. Tegra is no chip of it’s own, it’s a platform. It makes use of several chips,and although the main improvement of Tegra is graphical improvements, their CPU will not be marginally faster than the newest ARM versions.

               It’s like the story between Pentium 3 and Pentium 4 processors all over, where one has a 3,or even 300Mhz higher clockspeed, in daily life it really does not amount to much!

               Benchmarks show that we are talking of less than 10% in some cases,though graphically the tegra outperforms most ARM chips.

            4. Sorry, but you’re incorrect.  First I’m comparing present ARM devices like the 45nm OMAP 4 in this article, which are Cortex A9 to the next Gen that are not only going to 28nm FAB but also going to Cortex A15 with A7.

               Meaning a major architectural change that stands to grant a significant improvement in performance and those always equate to more than clock speed difference in performance.

               So don’t confuse them with simple FAB improvements that usually only allow increased clock speeds but under the same architecture.

               Second, these next gen chips will operate with clock speeds up to triple what is typical now.  The combination of going from 45nm to 28nm and the up to 5x power efficiency improvement (A7) and the growing demand for higher performance solutions from ARM means much faster clock speeds will be made available.

               Remember, these next gen ARM chips will be the first to actually run a major desktop OS with Windows 8 for ARM and will not be just another mobile OS platform anymore.

               Third, as already mentioned the Tegra 3 is only a intermediate.  It’s still based on Cortex A9 and is presently only reduced to 40nm.  So don’t confuse it with the next gen ARM chips actually being discussed.

               While the ATOM is also becoming a “Platform”.  Medfield is already a SoC and Intel is on its way to do so to it’s entire mobile line.

               Haswell for example will be the SoC version of Ivy Bridge.  While all ATOM chips will be getting a full architectural reboot with the 22nm Silvermont.

               And don’t think AMD is the only one developing APU like enhancements.  CPU and GPU are slowly becoming integrated under all the major chip makers.

               Understand, I still think Intel stands a strong chance of making enough improvements to compete with ARM.  I’m just pointing out where ARM is now and how long it’ll be before we see Intel really start competing in the mobile markets.

            5. First of all, sorry I meant ARM instead of AMD.

               This discussion seems to go in circles. So I’ll just summerise my point and won’t comment any further. By current generation of ARM chips I mean something like OMAP4. By the next generation I mean something like OMAP5. Saying that OMAP5 will be 5x faster than OMAP5 is absurd. Of course one can pick a SoC with 5x lower performance than OMAP5, but that is pretty pointless.

            6. You seem to compare the low end of current AMD chips (that is 1GHZ Cortex A8, this is actually the previous generation) to the speculated high end of the next generation (that is 3GHZ 8 core Cortex A15, I don’t think we are quite there yet). It would be fare to compare say OMAP4 with OMAP5.

            7.  Corrections… One, AMD doesn’t make ARM chips.  Twon, I specifically stated Cortex A9!

               And OMAP 4 to OMAP5 is comparing Cortex A9 45nm to Cortex A15/A7 28nm.

               While comparing to Cortex A8 would be a even bigger difference.  Since most were single core and clocked even slower than 1GHz on average.  In fact there are still low cost phones sold with 600MHz parts.

               Many of the Cortex A15 28nm products will be out before the end of the year and the rest next year.  So we aren’t that far away from them.

               Besides, we have to wait for next gen Intel ATOMs as well that won’t be out till next year as well before we see the real competition.

Comments are closed.