POST A COMMENT

92 Comments

Back to Article

  • Flunk - Thursday, April 23, 2015 - link

    ARM's core names are utterly bewildering, their marketing department should be shot. It they really wanted differentiation why don't they have a A31 for low end, A51 for mid range and A71 for high end? Then they have names for the next 8 generations ready to go. Reply
  • DCide - Thursday, April 23, 2015 - link

    Fortunately they don't market to consumers, for the most part.

    That's the only way they can get away with it. But you're still right.
    Reply
  • Bryf50 - Thursday, April 23, 2015 - link

    Don't forget the A17 which is a revised A12 which is still slower than the A15. Reply
  • Andrei Frumusanu - Thursday, April 23, 2015 - link

    Actually the improvements in the A17 over the A12 do make it outperform the A15. Reply
  • Onsager - Thursday, April 23, 2015 - link

    Agreed. Shot may be a bit harsh. But I agree. Reply
  • Ammaross - Thursday, April 23, 2015 - link

    He just said "shot," not necessarily where. Can still be a suitable punishment to kneecap them. Reply
  • FlushedBubblyJock - Friday, April 24, 2015 - link

    This is the only time the cattling cantanko crew seems to feel they and the rest of humanity aren't bright enough to see past a few numbered names.

    I find that summarily convenient.
    Reply
  • jimjamjamie - Friday, April 24, 2015 - link

    The barbarian in me would suggest that the threat of a NERF shooting should keep the employees in line. Rule with a rubber-tipped foam fist. Reply
  • Buk Lau - Thursday, April 23, 2015 - link

    LOL exactly what I thought. Their past naming schemes have been just random numbers with no clear pattern Reply
  • FlushedBubblyJock - Friday, April 24, 2015 - link

    They should massive rebrand for 5 years like AMD does. Reply
  • FlushedBubblyJock - Friday, April 24, 2015 - link

    That's the point, keep everyone confused and incapable of making a proper decision, and keep all the consumers befuddled, that way they can sell a racing stripe at a premium and by the time the ripoff sinks in it's too late the next con job has already hit - rebrand the prior and addd it to the current mix to confuse and obfuscate ever more.

    Looks like they deserve a Nobel Prize, not execution, rather for execution.
    Reply
  • mczak - Thursday, April 23, 2015 - link

    Hmm something doesn't add up with maximum possible L2 cache size. The table lists max 4MB for A15, and max 2MB for A57/A72, however the slide at the bottom shows up to 4MB for the A72 again. Reply
  • Andrei Frumusanu - Thursday, April 23, 2015 - link

    Thanks for catching that, corrected. Reply
  • mczak - Thursday, April 23, 2015 - link

    Are you sure it's right that way? I was merely pointing out the inconsistency. But actually looking at the arm site their diagram shows 2MB as well (http://www.arm.com/products/processors/cortex-a/co... so maybe the diagram was just wrong.
    (I find it somewhat interesting in any case since at least for the A57 it has gone down from 4MB the A15 could do to 2MB, especially considering that just about everybody used the max 2MB size for the A57 and noone bothered with less.)
    Reply
  • mczak - Thursday, April 23, 2015 - link

    Actually 4MB should be correct - don't trust the marketing material, trust the tech docs, and these indeed say 4MB for the A72. Reply
  • jjj - Thursday, April 23, 2015 - link

    I assume you are already aware but avoid to point out that there are some early Geekbench scores for the 2xA72+2xA53 tab SoC from Mediatek. Target clocks were 2.4 GHz but we don't know if the tested SoC reached that yet and ofc the software is evolving and newer results keep getting better.With that in mind http://browser.primatelabs.com/geekbench3/compare/...
    1700 score in single core is pretty exciting (even more so for a mid-range tab SoC) and likely the final perf will be at least a little better. On more advanced processes (this is on 28nm) we'll get slightly higher clocks and a boost on the memory side (DDR4, wider bus).
    Not long ago there was a Huawei slide comparing die sizes for A53, A57 and A72.
    A53 core 1.2mm2, quad clister 8.4mm2
    A57 core 3.6mm2 , quad cluster 20.7mm2
    A72 core 3.3mm2 , quad cluster 18.7mm2
    The first 2 were clearly on 28nm and i assumed the A72 is on 28nm too and since this article confirms a size reduction, now it's certain that it was on 28nm (likely TSMC).
    So A72 seems to be delivering and things should get exciting soon.

    PS will be pretty funny to see SD620 vs SD810 in CPU perf.
    Reply
  • Buk Lau - Thursday, April 23, 2015 - link

    I will laugh so hard, and it's probably going to happen that SD620 smokes 810 completely. Qualcomm pretty much gave everyone else a chance to catch up when it stopped using custom cores Reply
  • Andrei Frumusanu - Thursday, April 23, 2015 - link

    Actually as far as I know, that SoC is running 1890MHz on its A72's. Reply
  • Andrei Frumusanu - Thursday, April 23, 2015 - link

    Correction, 1989MHz. Reply
  • jjj - Thursday, April 23, 2015 - link

    lol that was shocking and took me a while to figure out that you might be right.
    I guess the final result is that high mostly on gains in the memory score while the gains in integer and floating point are some 10-12% over Exynos7420. So if they manage to hit the announced 2.4GHz, the Geekbench scores will be bananas but maybe too much of that will the memory score. The 1700 score does seem to match (more or less) the ipad air 2 in integer and memory but needs a 20% boost in FP.
    Impressive anyway, assuming A72 hits it's power targets, or at least close to that.
    And thanks for sharing that, reminded me to look beyond the total score.
    Reply
  • sonicmerlin - Friday, April 24, 2015 - link

    Slightly higher clocks? Going from 28 nm to 18 nm should result in a huge improvement. Reply
  • sonicmerlin - Friday, April 24, 2015 - link

    Oops I meant 14 nm. Reply
  • jjj - Friday, April 24, 2015 - link

    As it turns out those results should be at 2GHz but the SoC is supposed to hit 2.4GHz when it starts shipping. ARM is targeting 2.5GHz on 16/14 and that's just slightly higher vs 2.4. Ofc those 16/14nm SoCs will use 4xA72 and will be targeted at phones while this is just dual and for tabs. Reply
  • melgross - Saturday, April 25, 2015 - link

    I don't know how exciting a score of 1700 single core really is, as Apple's A8x does 1800 single core now. I hope that actual core speeds for real production chips would be at least 20% higher. Otherwise Apple's A9 will glide right past. Possibly a new Tegra might as well. Reply
  • Mashrur - Thursday, April 23, 2015 - link

    Just hoping A72 doesn't get end up like Intel netburst. Reply
  • MrSpadge - Thursday, April 23, 2015 - link

    A72 is pretty much the opposite of Netburst. Don't worry about it bursting anything but benchmark records in its class ;) Reply
  • jospoortvliet - Friday, April 24, 2015 - link

    Because of a 2.5ghz max speed? I think that's quite ok in a mobile device, Qt least as burst speed. Sustained, 1.5 ghz will probably be doable on 14nm perhaps even a bit less. Reply
  • joex4444 - Friday, April 24, 2015 - link

    Netburst was a deliberate attempt to win the GHz race by increasing the pipeline. Notice the A72 has a maximum pipeline length of 16 cycles versus the A53/A57's 19 cycles.

    It's anti-netburst.
    Reply
  • Novacius - Thursday, April 23, 2015 - link

    They have finally fixed the broken interconnect and/or memory controller, thus a 50% gain in memory performance. Reply
  • Andrei Frumusanu - Thursday, April 23, 2015 - link

    We've addressed ARM's memory performance numbers in our review of the A57/Exynos5433/Note4 article and the reasoning why it (misrepresentatively) gets the numbers it gets. The increases on the A72 should be Independent to the interconnect and memory. Reply
  • Novacius - Thursday, April 23, 2015 - link

    Thank you. I didn't knew that. Reply
  • Frenetic Pony - Thursday, April 23, 2015 - link

    After the utterly drivel performance of the a57 I'm not even interested. Looking forward to details of Qualcomm's 820 and AMD's Zen Core, as they might actually have interesting performance gains/be in devices I'd actually end up buying. Reply
  • Novacius - Thursday, April 23, 2015 - link

    The A57 itself performs rather good. Just have a look at the Exynos 7420. Reply
  • Frenetic Pony - Thursday, April 23, 2015 - link

    That's built on Samsung's new 14nm node, and that's where all of the performance gains come from. The chip itself, on 28nm, underperforms Qualcomm's Krait cores in every way. Reply
  • lopri - Thursday, April 23, 2015 - link

    ARM's big cores (A15, A57, A72) have higher IPC than Krait cores. The 7420 has higher frequency and better memory, as well as whatever optimization Samsung did, but at the end of the day it is still an A57. Reply
  • Frenetic Pony - Thursday, April 23, 2015 - link

    Yet the performance per watt, which is what matters, of an a57 core at 20nm node is little to no better than a Krait core at 28nm despite the process improvement. So, I repeat, a57 cores are junk. Reply
  • Frenetic Pony - Thursday, April 23, 2015 - link

    For clarification, the jump in performance over a variety of factors from 28nm which Krait is built on to 20nm Finfet or "14nm" as Samsung calls it, is quite big. And if Krait cores made the same jump I wouldn't be surprised if they beat out the a57 cores. In recent times ARM's stock cores have had a history of being handily beaten by major ARM licensees, and the a53 and a57 cores don't seem any different. Nor does the a72 core, as the claims vs actual performance of the a57 were so far apart that the even less claims of a72 improvement are still easy to dismiss.

    Qualcomm on the other hand had the entire top end, except for Apple, for the past few years with its Krait cores. And there's little reason to suspect, so far, that their Krait successor, late as it is, will be any different. Unless of course AMD manages to pull a company saving CPU out of their hat, but the designer of Zen designed their last big CPU hit, the one that made them beat Intel for a while. So, perhaps, he can do so again.
    Reply
  • lilmoe - Friday, April 24, 2015 - link

    You need a lot of experience with ARM's big.LITTLE reference architecture to minimize the shortcomings. Samsung has a lead in that regard over Qualcomm, so you also need to put that into perspective. Mediatek and HiSilicon aren't even touching the A57 in their big.LITTLE configuration.

    Samsung has somewhat solved/minimized these in roundabout ways (Exynos 5433/7420). The A72 and CCI500 might be addressing these very same big.LITTLE issues when it comes to switching, efficiency and sustained performance, so I'm guessing we won't be seeing *huge* improvements in overall platform performance in _Samsung's_ case, but that's definitely not the case for Qualcomm and others.

    The only realistic comparison would probably be if Samsung releases A72 on their 14nm process and compare it to the 7420, this way Samsung's own optimizations would be factored in. The problem is that rumor has it that Samsung might be releasing their own custom cores starting next year and the Note 5 will probably be still rocking A57's...
    Reply
  • Impulses - Thursday, April 23, 2015 - link

    A72 cores on the next Nexus at year's end? Reply
  • hammer256 - Thursday, April 23, 2015 - link

    So, this might sound cynical, but doesn't these gains in the A72 over the A57 suggest that the A57 wasn't really well designed to begin with? After all, it's not easy to get higher IPC, less power, and less area(?) on the same node if the previous design was well optimized to begin with.
    But, I'm going to try not be cynical.
    Reply
  • mczak - Thursday, April 23, 2015 - link

    Well, time to market. Earlier chips are always going to be less optimized than later similar tweaked ones appearing later.
    Just look at Cortex-A8, if you compare this against Cortex-A7 it is complete junk. Something like twice the die size (at the same process node), but slower (ok similar performance actually), more power hungry, and missing tons of features the Cortex-A7 has (like being MP-capable etc.). But I wouldn't exactly call it a failure :-).
    Reply
  • Alexvrb - Thursday, April 23, 2015 - link

    Agreed. A57 may have been a stopgap solution, but they needed something to launch sooner - and so did the phone manufacturers. Better to release A57 now and A72 later, than to simply do nothing until A72 rolls around (and piss off big customers who are on a schedule). A57 is certainly better than A15.

    I'm really looking forward to seeing what the next custom Qualcomm core can do.
    Reply
  • hammer256 - Thursday, April 23, 2015 - link

    Yeah, you are right. Maybe that time to market pressure was put on by Apple's designs? Reply
  • psychobriggsy - Friday, April 24, 2015 - link

    You have to remember that this was ARM's first high-performance (for mobile) 64-bit core. Therefore there were bound to be many areas that could be improved, and I'm sure that internally the A72 refinement process has been going on during the A57 core's final stages of completion.

    ARM has a history of having multiple revisions of a core design. You'll sometimes see them referenced, e.g., Cortex A15 Revision 3 was used in a Tegra design. These historically improve power consumption, fix bugs, increase performance, etc. This A72 looks to me like a very impressive core revision that's warranted a new marketing name.
    Reply
  • tsk2k - Thursday, April 23, 2015 - link

    Thanks for the article Andrei.
    What I'm wondering is if qualcomm will be able to make the snapdragon 820 faster than the snapdragon 620, any thoughts on this?
    Reply
  • jjj - Thursday, April 23, 2015 - link

    Any clue at all how the cache is scaling on 16/14nm?
    If it scales poorly it could reach 50% of the quad cluster die area,it wasn't too far in the Exynos 5433 at 46%. It's somewhat strange that cache is using so much die area and not sure how much sense that makes. At this rate Renesas and Cypress will make SRAM with on die CPUs soon not the other way around lol.
    Reply
  • hammer256 - Friday, April 24, 2015 - link

    Huh, that's weird. So the cache die area isn't shrinking proportionally as other parts? Why would that be? The interconnects? Reply
  • jjj - Friday, April 24, 2015 - link

    SRAM doesn't scale all that well.
    As to why, there are a bunch of factors but i'll let someone more capable on the matter explain it, or you can try to search for some info.
    Reply
  • Morawka - Friday, April 24, 2015 - link

    SRAM Scales fine but there are trade-off's. see 10nm SSD NAND Reply
  • Xtradi - Thursday, April 23, 2015 - link

    "...in other power optionizations to the buffers..." What is optionizations ? Reply
  • Ryan Smith - Friday, April 24, 2015 - link

    It's like an optimization, but it's optional. Reply
  • LukaP - Friday, April 24, 2015 - link

    Interesting. I need to write this one down... Reply
  • Drumsticks - Friday, April 24, 2015 - link

    No thoughts on the ambitious goal/stated intention of using 75% of the power of broadwell for similar performance? Arstechnica indicated that ARM felt VERY confident in that claim. Reply
  • iwod - Friday, April 24, 2015 - link

    I am interested in how it compares to 14nm Intel Atom Core.
    I am going to assume Intel still has the edge on performance? Or has that finally been closed down with A72?
    Reply
  • iwod - Friday, April 24, 2015 - link

    I am interested in how it compares to 14nm Intel Atom Core.
    I am going to assume Intel still has the edge on performance? Or has that finally been closed down with A72?

    Edit: Turns out that Ars has a slide that said A72 is 75% within Broadwell, truly impressive.
    Reply
  • jjj - Friday, April 24, 2015 - link

    Atom has never had the edge on perf and the 14nm Atom is just a die shrink without higher clocks so A72 obliterates it but to be fair i have no idea how big the Atom core is on 14nm and we are yet to be certain that A72 hits it's power targets.
    As for the Broadwell slide it seems to be Broadwell at 2GHz vs A72 at 2.5GHz and the under 1W vs 4W is a bit misleading. Doubt a single core will use all those 4W.
    http://cdn.arstechnica.net/wp-content/uploads/2015...
    In multi core 4 ARM cores would be at 3W(4x750mW) vs Intel at 4W with dual core /4 threads and the cores would fail to reach the max 2GHz because of the 4W limitation.
    Single core vs single core A72 at 2.5GHz might match Broadwell at 2GHz in integer and memory but not in FP, at least that seems to be the case in Geekbench.
    Those benchmarks in the slide are not fake, they are just slightly creative so you have to avoid the traps. Also ARM doesn't actually have the silicon for A72.
    In 4 cores vs 2 cores and likely even 2 cores vs 2 cores at the tested speeds, A72 beats it when Broadwell is limited to such low power.
    Ofc Broadwell is not designed for low clocks and low power and it's a hell of a lot bigger so it's not all that impressive in this kind of scenario.
    A72 expected perf is very exciting and things will get more interesting.
    A few days ago a slide leaked with future ARM cores and it seems they will have a core called Ares on 10nm targeting 1 to 1.25W while A72 is targeted at 0.75W. A 66% increase in power and a new process should mean big perf gains.
    Reply
  • Speedfriend - Friday, April 24, 2015 - link

    @jjj
    Even with the usual grain of salt when looking at comparatives that ARM slide is really pushing it.

    It is comparing the lowest Spec Broadwell 5Y10 which has a geekbench single core score 40% lower than the top spec 5Y71.

    As there is no single core broadwell and there certainly won't be a single core A72, comparing the full power consumption of Broadwell at 4W against A72 at 1W is just plain lying.

    They are making the assumption that the Core M will be thermally limited and not achieve full target frequency, when Anandtech results have shown that it is highly dependent on the work load and form factor.

    It is comparing a theoretical design potentially available early next year, with a shipping chip already available for 6 months and which will have already been replaced by a chip likely to have 20% better performance.

    I now know why the guys at Imagination tend to be highly sceptical of anything that Arm claims.
    Reply
  • psychobriggsy - Friday, April 24, 2015 - link

    In the end the slide is making the point that the A72 achieves similar results in a lower power budget than the Core M, whether it's a single core or multiple cores. When the chassis cooling is good, then Core M can turbo beyond its TDP for longer, so is it still 4W? How would A72 perform if allowed to run at a higher TDP?

    In addition, SoCs using A72 are going to be priced in the $40 range at most, possibly down to $20. Core M is in a different ballpark in this regard.

    However Core M is available now, whereas the first A72s will be late this year at best.
    Reply
  • jjj - Friday, April 24, 2015 - link

    My estimates for Geekbench were based on existing results for the MT8173 and those are in line with what i said and what ARM said.Plus i did my part in trying to explain how ARM is being "creative"
    Ofc while we have some early perf numbers for a dev board ,we don't have any real world power numbers.
    I will also point out that he quoted price for Core M is 281$ while dual A72+dual A7 tablet SoCs will likely be as low as 10$.
    Amazon is using 2xA15+2xA7 from Mediatek in it's cheaper tabs and this dual A72 is replacing that so soon enough we might see it in some 100-150$ Kindle.
    Reply
  • V900 - Friday, April 24, 2015 - link

    You do realize that there isn't any reason beyond appetite for profit margin, that Intel charges 280$ for a Broadwell CPU, right?

    Now IF ARMs theoretical design comes out in a year, and will offer close to the same performance as Intel's slowest Broadwell processor that came out six months ago, there's nothing that prevents Intel from smacking an Atom label on them, and sell them for 40$ or 20$.

    (Actually that's probably what they'll do, since Intel will have a much faster Skylake or Cannonlake CPU out by then to charge 200-400$ for...)

    Why do people often treat ARM like some chubby kid with Down's syndrome who's competing at a track meet because nobody wanted to tell him he didn't make the cut?!?

    "Oh wow, look at that ARM-kid! He's almost as fast as the Intel runner! GOOD JOB ARM!! WHOOO YOU ROCK!! You know, he could actually win this if he had a 1 lap head start!"
    Reply
  • sonicmerlin - Friday, April 24, 2015 - link

    Intel is not going to rebrand their Core M chips and sell them as Atoms. Not only would that be utterly stupid from a financial perspective (why would anyone buy a Skylake chip if they could get a Core M with similar performance for a tenth the cost? And why do you think Skylake will have such improved performance when Intel hasn't improved their performance in 5 years?), but ARM's chips are running at a much lower TDP than Intel's. That's why ARM gets the praise. Reply
  • name99 - Friday, April 24, 2015 - link

    You think in a YEAR Intel will be selling Cannonlake? Good luck with that prediction.

    As for "much faster Skylake", apart from AVX-512, what makes you believe that? The recent history of Intel has been maybe 15% speed increases at major new architectures (from Nehalem to Sandy Bridge) and around 5% for the subsequent tweaks, eg Sandy to Ivy to Haswell to Broadwell.
    Even if Skylake falls into the 15% category, that's hardly "much faster".
    Reply
  • jjj - Friday, April 24, 2015 - link

    Why do you treat Intel like they don't utterly suck? Or like greed is a virtue?
    Intel can't make a competitive Atom, it's always next year
    And maybe you should factor in die sizes.
    A72 is 3.3 mm2 on 28nm, bellow 1.9mm2 on Samsung 20nm and some 1.6nm2 on Samsung 14nm. You have any clue how huge Broadwell is, some 4 times bigger than A72 on 14nm. So great Intel has a core that costs 4 times more, they deserve a Nobel in economy.
    Intel can sell Broadwell at 40$ sure since it's 82mm2 while the ARM guys can do A72 at 10$ this year on 28nm and well bellow that later on smaller processes.
    Sure the Braadwell core is not made for this kind of load and there is nothing wrong with that but you don't get to claim that it can compete with a core that's a few times smaller in the real world.
    Reply
  • name99 - Friday, April 24, 2015 - link

    OK. Call us when Core-M gets its first phone win.
    Until then, ARM's comparison is perfectly legit.
    Reply
  • psychobriggsy - Friday, April 24, 2015 - link

    In terms of comparison with Atom, A9 was 'competitive' in terms of performance with the Atom cores of the time. A15 often exceeded (significantly). A57 should be in front of current Atom cores, and this should be further still. Intel had a process lead (used to reduce power consumption and have good turbo speeds), but that's reduced now with 14FF from Samsung (sure, there are 20nm elements still, but it's not a full node or two advantage for Intel now). Reply
  • mczak - Friday, April 24, 2015 - link

    The Cortex-A15 never really exceeded performance of the (silvermont) atom, at least not significantly. Even though based on paper specs, it should have (as it's a wider design). Ultimately though in the form factors we're talking about performance was limited by power, and the A15 didn't seem to have an edge there.
    Likewise, the Cortex-A9 should have outperformed the then current slow bonnell core atom, but likewise didn't really (the execution core was definitely faster, but the atom had the edge in things like branch prediction, memory pipeline, overall this pretty much was a wash).
    But the A72 (when using a finfet process) should have a good chance of outperforming the atom I guess (since the new 14nm atom didn't improve all that much). I don't know if it's more energy efficient, but peak performance at least should probably be somewhat higher (as it should reach similar clocks but be faster per clock). I don't think though the differences will be too great (it's always difficult to tell due to the different architecture used, hence even when using the same software compilers etc. can have quite some influence).
    Reply
  • Wilco1 - Saturday, April 25, 2015 - link

    Just like the A9 outperformed Silverthorne, Cortex-A15 outperforms Silvermont by a huge margin:

    http://browser.primatelabs.com/geekbench3/compare/...

    To get an idea how Atom will do compared to A72, here is how Galaxy S6 compares with the fastest Atom running Android:

    http://browser.primatelabs.com/geekbench3/compare/...

    Yes that shows A57 is ~80% faster clock for clock than Atom. Now consider that A72 will do 2.5GHz and is 16-30% faster per clock as well, so it is safe to say A72 will be at least TWICE as fast as Atom when it comes out in a few months.

    So A72 "has a good chance to outperform Atom" is the understatement of the year...
    Reply
  • 68k - Saturday, April 25, 2015 - link

    I get the same relative performance in Geekbench when comparing Silvermont vs Cortex-A15, so my boards are not obviously broken. Yet, when running the same Debian version on both boards, I cannot find any non-trival "real" program where Silvermont is at least as fast as Cortex-A15 clock-for-clock. Silvermont beat A15 with a huge amount in programs with a lot of random memory accesses in data set that far exceed the size of the cache.

    Maybe Geekbench isn't the most reliable measurement of general purpose performance, especially not when comparing result from different ISA.
    Reply
  • Wilco1 - Monday, April 27, 2015 - link

    Geekbench tracks real world performance reasonably well and gives similar results as SPEC when comparing different CPUs. It is not a random memory test of course, but it does exercise L2 and main memory, so having a good memory system and prefetchers helps a lot.

    Since you mention dev boards, in my experience ARM dev boards are lacking in memory performance (both latency and bandwidth) - performance in real devices tends to be much better. As an actual example a dev board I used gives less than 50% of the bandwidth of my phone when running the same benchmark. For Intel boards it may be the other way around - for example the latest 22nm Atoms are slower than the initial Silvermont dev boards - look at the memory test, especially multithreaded:

    http://browser.primatelabs.com/geekbench3/compare/...

    So when comparing performance devboards are not necessarily representative.
    Reply
  • mczak - Saturday, April 25, 2015 - link

    You use the Tegra K1 for comparison, which isn't really very representative overall. This uses a later revision of the Cortex-A15 than most, and nvidia also managed to clock it higher (it is of course also probably the newest SoC still using a A15). That alone is good enough for roughly 20% more performance than what your typical Cortex-A15 SoC provides.
    Besides, geekbench isn't quite that indicative of real world performance. Yes, the execution core of the A15 is faster than Silvermont (just like Cortex-A9 was faster than Silverthorne), as it should be, but that did not really translate to any real world advantage, not least because the memory pipeline of the atoms was superior, which isn't really reflected in geekbench scores.
    There's imho no way the A72 will be twice as fast in any kind of real world task. If it reaches 50% faster that would be quite an achievement already.
    Reply
  • Wilco1 - Monday, April 27, 2015 - link

    Well the Exynos 5422 in the S5 still beats the very latest Atom by a good margin despite its age and lower clock. Something like ASUS MeMO Pad 7 would be almost comparable (about a year ago, small tablet vs big phone, both 32-bit Android), and that can only burst to 1.86GHz, so it ends up a good deal slower than the Exynos.

    There is no doubt the early A15's had issues with the memory system, but later revisions improved significantly (and newer cores have much better memory systems).

    Clearly no one benchmark is perfect, but without good real-world benchmarks we've got to use what is available. And much of people's perception of what is CPU performance in actual devices is incorrect, for example smooth scrolling and zooming is all about software using the GPU, fast app launching depends on flash, JS performance is mostly software etc.
    Reply
  • V900 - Friday, April 24, 2015 - link

    Also, Intel never had a process advantage on Atom, since they only made them in their old fabs, which were always around two full nodes
    Behind: 32nm instead of 22nm...
    Reply
  • beginner99 - Friday, April 24, 2015 - link

    Am I the only one taking this 3.5 figure with a huge grain of salt? I don't believe it especially after seeing A-57 should already be 1.9x as fast as A-15 and more power efficient which we already see isn't really the case. Snapdragon 810... Reply
  • FlushedBubblyJock - Friday, April 24, 2015 - link

    Yes, take the worst case scenario throttling on the old 15, combine it with the best case scenario unachieved highest sustained and boosted mhz and throughput theory on the new 72, and lo and behold 350% !!!!!!!!!!!!!

    I'm sure the dress is very beautiful.
    Reply
  • heartinpiece - Friday, April 24, 2015 - link

    "When on the 28nm node, we see the A72 having a respectable 20% power reduction when compared to the A57. As a reminder - we're talking about absolute power at the same clock speed"

    Doesn't that slide show the 'energy consumed?' The A72 probably executes faster, thus the energy delay product is probably lower. I don't think the slide shows a power consumption comparison.
    Reply
  • FlushedBubblyJock - Friday, April 24, 2015 - link

    It's "not clear", that way, the claims are just an "impression" that can be "interpreted" and thus remain free from "lies" aka lawsuits.

    Thank you for your confusion, please attend the next group impressions session as the perfect guinea pig.
    Reply
  • darkich - Friday, April 24, 2015 - link

    Going by these performance comparison numbers, it looks like A72 will be sensational and pretty much outperform, for example, the current Core M at a fraction of the price and half the TDP.

    Up to 50% higher performance than A57 on the same node and clock,yet unrestricted at 2.5Ghz..wow!
    Reply
  • ShieTar - Friday, April 24, 2015 - link

    How do you figure that?
    If I understand the designs correctly, a Haswell-Core with his 5-cycle FMAC and 2x256bit ALUs should be able to reach 240% of the performance of a A72-Core with a 6-cycle FMAC and 2x128bit ALUs, at the same frequency.
    Reply
  • darkich - Friday, April 24, 2015 - link

    If you understand the designs correctly, compare the AMD Fx Core with an i3 and tell me which one should be faster.

    My figures come from the respective single core Geekbench scores of 14nm 2GHz A57, and mid range Core M.
    2.5Ghz A72 should score around 2000 and, more importantly without turbo tricks. In other words, it should have superior sustained performance
    Reply
  • 68k - Saturday, April 25, 2015 - link

    Yeah, no "turbo tricks" because high frequency ARM designs always run at maximum clock and never throttle...

    Cortex A72 look like a very good product, but it ridiculous to compare it to Core M. Completely different price point and the aim at different but somewhat overlapping markets.

    One of the most important aspects of modern big-core designs is the memory system and Intel's big-core design far exceed ARM at this. Geekbench seem to be useless to test the memory system aspect of a CPU, the "memory" result only does some very basic bandwidth testing. What matters is average latency on memory operations.
    Reply
  • darkich - Saturday, April 25, 2015 - link

    I am comparing the purely from the technological perspective. And from that point, what ARM achieves is by far more impressive than the best efforts of the so called Chipzilla. Reply
  • V900 - Friday, April 24, 2015 - link

    Wow! I must say my good man, you sure do have an eye for detail, an intuitive grasp of numbers, and a quick mind that most of us can only envy you... I tip my hat to you, good sir!

    And may I tell you something in confidence? I represent a group of businessmen, who are looking for discerning, smart and worldly critical thinkers, such as yourself, for a unique and highly profitable investment opportunity in Florida... May I interest you in a prospectus?
    Reply
  • darkich - Saturday, April 25, 2015 - link

    Not sure if that's an irony or..and not sure to whom you are replying to.
    Anyway, I had fun and a good laugh reading the comment. Thanks, kind sir.
    Reply
  • blanarahul - Friday, April 24, 2015 - link

    "When on the 28nm node, we see the A72 having a respectable 20% power reduction when compared to the A57. As a reminder - we're talking about absolute power at the same clock speed"

    Nope. We are talking about absolute power at same "performance". This is why they mention that for obtaining equivalent performance you run A15@1.6 GHz, A57@1.3 GHz and A72@1.1 GHz.
    Reply
  • toyotabedzrock - Friday, April 24, 2015 - link

    Problem with the a57 was that it was just an a15 with the needed 64 bit changes. They rushed out the a57 which is why the a72 is coming so soon. Reply
  • FlushedBubblyJock - Friday, April 24, 2015 - link

    So it's really a 25% poweer savings, from 20nm, but for marketing lies, they claim 75%, going from 28nm....

    TWENTY FIVE PERCENT POWER SAVINGS AT BEST
    Reply
  • toyotabedzrock - Friday, April 24, 2015 - link

    The a57 was a 64bit extension of a15 this is a new core which is what they had been working on.

    I would like to have seen some comparison to the issue width of the apple and nvidia cores.
    Reply
  • jerrylzy - Friday, April 24, 2015 - link

    ARM is probably comparing with the earliest version of Cortex A15. Reply
  • name99 - Friday, April 24, 2015 - link

    "The obvious partners that might ramp prodution the soonest are MediaTek and Qualcomm,"

    Isn't QC's Kyro (or whatever it's called) ready to roll soon? Seems like they'd go with that rather than the A72, especially given the hash they have made of the A57 in the 810.
    Reply
  • jjj - Friday, April 24, 2015 - link

    Qualcomm announced it's new midrange SoCs with A72 at 1.8GHz, the SD618 and SD620. Reply
  • Wardrive2015 - Wednesday, April 29, 2015 - link

    Any idea on how large the reorder buffer is? Anyone know how large it is on A57? All I can find is a statement saying greater than 128 for the a57, but how much greater? Anyway the A72 looks like its going to be a beast of a core. Its amazing they got greater performance, less heat and even less area than the A57. I hope we get a next generation Tegra with this core, and that they drop Denver, or have 2 versions Reply
  • Wardrive2015 - Wednesday, April 29, 2015 - link

    Oh and the Cortex A15's NEON units are 128 bit, not 64 bit. Reply

Log in

Don't have an account? Sign up now