Arm Says Future Chips to Offer Desktop-Class Performance

Posted on August 16, 2018 by Paul Thurrott in Hardware, Mobile with 27 Comments

Arm today disclosed its roadmap for CPU performance improvements through 2020. And the chipset designer has Intel in its sights.

“Advances in Arm technology have brought desktop-class PC performance into our smartphones, fundamentally changing how we use technology in our daily lives,” Arm’s Nandan Nayampally explains. “Arm is now applying this same design leadership … to enable the PC industry to overcome [its] reliance on Moore’s law – which has definitely slowed – and deliver a high-performance, always-on, always-connected laptop experience.”

Put simply, Arm—like Qualcomm, its biggest licensee—is pushing into the PC market and is advancing its chipsets for this market at an accelerated rate. The results are smaller, more efficient chips—10 nm today, 7 nm later this year, and then 5 nm in 2020—and, as important, more powerful chips.

Whether Arm’s chipsets will ever truly compete with Intel is sort of beside the point, as in some ways they don’t have to. With Intel pushing forward with more powerful chips that have more processing cores, it will have the premium and gaming markets to itself. This leaves the larger, mainstream part of the market open to some competition. And Arm—and Qualcomm—may be ideally positioned to compete there.

“Our client CPUs … combined with innovations from our silicon and foundry partners will enable Arm SoCs to break through the dominance of x86 and gain substantial market share in Windows laptops and Chromebooks over the next five years,” Nayampally claims.

That said, Arm’s comparisons are a tad disingenuous. It is claiming that its 2019- and 2020-era chips will outperform Intel’s Core i5 chips. But the Intel chips it is comparing are previous-generation dual-core designs, not the quad-core 8th-generation chips that Intel started shipping last year.

Which is fine, on some level: After all, dual-core Intel Core chips offer enough performance for virtually all mainstream users.

So the real question is whether Arm and licensees disrupt the PC/Chromebook market. And on that note, yes, I think they can. And some combination of Core i5 performance—even previous-gen Core i5 performance—with incredible battery life and integrated connectivity is the right recipe for the mainstream part of the market.

 

Tagged with ,

Join the discussion!

BECOME A THURROTT MEMBER:

Don't have a login but want to join the conversation? Become a Thurrott Premium or Basic User to participate

Register
Comments (27)

27 responses to “Arm Says Future Chips to Offer Desktop-Class Performance”

  1. glenn8878

    ARM Mobile devices has to compare with Intel Desktop performance. It was never a equal comparison. Thus, whether ARM will ever catch up is almost academic since Intel will never get more efficient. As long as Qualcomm doesn’t give up, in a few generations, more will convert over unless Microsoft bungles Windows on ARM as they did with Phones.

  2. red.radar

    Integrated connectivity is a buzzword. And it doesn't matter because people do not want to pay a cell provider a monthly device charge to use their laptop on a capacity limited dataplan. This is a solution looking for a business model. Wifi is pervasive and free of bandwidth limits ect... Its also more power efficient and reliable.


    This is not going to be succesful unless they partner with the Telecom companies to give away connectivity. The only way they win is through power efficiency and proving they have good enough or equivalent performance to Intel with superior battery life. This has nothing to do with connectivity. It is about performance.


    What makes this interesting is Intel seems to have dropped the ball on their manufacturing nodes as they still can't get 10nm out the door..


    also.... lets not overlook the elephant in the room. Who is cross compiling the Win32 apps for Arm ?


  3. Stooks

    For any of this to be successful one of two things would need to happen.


    1. ARM licenses x86/64 and builds that capability into ARM chips.
    2. Everyone ports their software to ARM because as we know, emulation SUCKS.


    I do not think either will happen but I am not sure which one is less likely. I highly doubt Intel would license x86 to ARM. Even if they did ARM chips would become way more complex and then they would run into the same 10nm issues that Intel if facing.


    Intel is a leading CPU maker. They have been ahead of the others for a very long time. If they are having issue with 10nm or less why does anyone think that other companies can do it better? Yes some others will reach 10 or 7nm but those are on way less complex chips. At the same complexity no one is doing 10nm yet. So far nothing beats Intel's IPC. AMD has closed the gap but they are about 7% slower over all and in some cases more than 10%.

    • skane2600

      In reply to Stooks:

      I agree. Regardless of license issues, Intel is still the best candidate for improving x86/64 performance. Intel is just serving the demand for x86/64 processors, it's not as if Intel couldn't design a non-legacy CPU just as powerful and energy efficient as an ARM chip.


      As far as number 2 is concerned, developers will have to see a very significant buy-in to ARM by desktop/laptop users before they're likely to port their applications. IMO improved battery life isn't going to be a sufficient motivator for ARM adoption to reach a tipping point.

    • Oreo

      In reply to Stooks:

      It is not true that the chips of the others are “way less complex”. Apple's ARM cores have had a comparable complexity to Intel's for several generations now by any measure (e. g. number of transistors). Secondly, software already has been ported en masse to ARM, because practically all smartphones and tablets run on ARM. The development ecosystem for ARM instruction sets is very mature and tons of money is invested into it.


      The fact that Intel is currently one, if not two process nodes behind the competition makes things even harder for them, because their chips run hotter and they have a smaller transistor budget.


      Secondly, because Intel has only a single core that stretches from 5W TDP to 125+ W TDP, their core designs are less optimized for either. The ARM ecosystem on the other hand has many different cores for many different TDP ranges, starting in the milliwatt range. Intel is lagging behind here, basically all top-of-the-line ARM chips use variations of big.LITTLE where they augment big, fast cores with more energy efficient, smaller cores. For mobile workloads this translates to more battery life. Intel doesn't offer that.


      ARM-based SoCs are also much more flexible when it comes to integrating other bits of silicon such as powerful graphics and a 4G modem. Again, Intel doesn't offer that customizability to anyone but very, very few select customers. And its integrated GPUs are no match for what AMD and the ARM ecosystem has to offer.


      How easy porting apps is is largely a function of how Microsoft manages the transition, not one of principal technical obstacles. Moreover, once a comparatively small number of select important apps has been ported, that is a non-issue. MS Office, for example, has already been re-engineered a few years ago to be portable, and the fruits of that effort already run on ARM (e. g. via the iOS versions of Office). Apple has managed several of these transitions very well, and there is no reason why Microsoft couldn't.

      • skane2600

        In reply to Oreo:

        "It is not true that the chips of the others are “way less complex”. Apple's ARM cores have had a comparable complexity to Intel's for several generations now by any measure (e. g. number of transistors). "


        As you state later in your comment, a lot of ARM chips integrate additional functions beyond the CPU. Those functions may require lower clock speeds and thus lower power. So although I'm not sure if the number of transistors matter either way, ARM vs Intel as far as complexity is concerned may be a bit of an Apples to Oranges comparison.


        "Secondly, software already has been ported en masse to ARM, because practically all smartphones and tablets run on ARM."


        I think it would be correct to say that a lot of mobile code has been designed specifically for ARM. I don't think it's correct to say that a lot of non-ARM based programs have been ported to ARM. Desktop class performance wouldn't mean much if it's not being used to run desktop class applications.

        • Oreo

          In reply to skane2600:

          As you state later in your comment, a lot of ARM chips integrate additional functions beyond the CPU. [...] So although I'm not sure if the number of transistors matter either way, ARM vs Intel as far as complexity is concerned may be a bit of an Apples to Oranges comparison.”

          It is not an Apples to oranges comparison, because the number of transistor determines the overall manufacturing complexity, die size and cost. And here many ARM SoCs have been on par with Intel's CPUs for quite a while. The Apple A8 from 2014 had ~2 billion transistors on a chip whereas a Broadwell-U from the same time period featured between 1.3 and 1.9 billion. The A11 that is currently state-of-the-art has 4.3 billion transistors.


          Moreover, if you compare the architecture of modern high-performance ARM cores, then they share the same architectural features as Intel's Cores do: out-of-order execution with very advanced predictors, 6+ wide execution, large caches, etc. That's why they are in the same IPC performance class as Intel and AMD CPUs.


          “Those functions may require lower clock speeds and thus lower power.”

          No, those other processors are gated differently, their clock speeds have nothing to do with the clock speeds of the CPU and GPU cores. In any case, I think it is the other way around: ARM cpus are designed with performance per watt in mind as the first priority, and they have different cores for different performance regimes and power envelopes. In contrast, Intel has only one. That's one reason why Apple's A-series chips smokes the competition, Intel and otherwise, because Apple is able to design chips with one specific performance and power target in mind. This has less to do with the extra logic, although that helps to keep power down, too.

          • alysdexia

            In reply to Oreo:

            "ARM cpus are designed with performance per watt in mind as the first priority, and they have different cores for different performance regimes and power envelopes. In contrast, Intel has only one."


            What? Xeon, Core, Pentium, Celeron, Atom: 5.


            W(CPU/(CPU+GPU)): best CPUmark 2019 Apr: /W, /$

            4.5 ~3/4: i5-7Y57, 3810: ~1128, 10  Kaby Lake

            5 ~13/16: i5-8200Y, 4206: ~1035, 14  Amber Lake

            15 ~6/7: i7-8565U, 9028: ~701, 22  Whiskey Lake

            7 ~6/7: i5-8210Y, 4057: ~676, 14  Amber Lake

            4: T5700, 2039: 510, 7  Apollo Lake

            28 ~9/10: i7-8559U, 12303: ~488, 28  Kaby Lake R

            6 ~11/13: N5000, 2443: ~481, 15  Gemini Lake

            35 ~48/55: i7-8700T, 12657: ~415, 25  Coffee Lake


            W(CPU/(CPU+GPU)): core–unit CPUmark [mobile/60] (Gn/s), Geekbench, UserBenchmark Int 2019 Apr: /W, /$

            4.5?: A12, [8103] (27), 4774–11480, : [1801?], ; 2551?, ; ,  Vortex-Tempest

            7.5? A12X, [12640] (45), 4993–17866, : [1685?], ; 2382?, ; ,  Vortex-Tempest

            2.4?: Cortex-A53/625, [2604], 870–4320, : [1085?], ; 1800?, ; ,  Apollo

            35 ~48/55: i7-8700T, 2351–12657, 4600–19000, 112–752: ~362, 25; ~540, 37; ~22, 1  Coffee Lake

          • skane2600

            In reply to Oreo:

            I never claimed any relationship between slower non-CPU element's clock speed and that of a CPU or GPU. I was just suggesting that if those elements share the same die as a CPU than the overall power requirements for the die might be less than an equally-sized die that is 100% dedicated to CPU functions even if the number of transistors is the same in both cases.


            Of course, it depends on a lot of architectural details that may vary from implementation to implementation, so it might not be possible to make a definitive general statement.

            • Oreo

              In reply to skane2600:

              I am not sure I understand what you are saying. These days no die is dedicated 100 % to CPU functions, all come with memory controllers and most also with GPUs. The caches, for example, take up a substantial amount of die area. The TDP is then shared across all elements. That means it is *harder* for Apple and other ARM SoC makers to be competitive with Intel in the CPU performance department, because their cores have to be more efficient. But benchmarks bear out that at least Apple A-series has been competitive for at least two generations.


              Also, we are making a mistake if you only look towards *C*PU performance, *G*PU performance is extremely important as well, and Intel isn't doing too well here either.

              • skane2600

                In reply to Oreo:

                Certainly what makes up what one might loosely call the CPU "system" has evolved over time to include those elements you mentioned, but I was referring to functions that aren't traditionally associated with the basic functionality of processing. Perhaps those functions aren't typically included in the same die, but I'm reaching the limits of my knowledge, so I couldn't say.

  4. DaveHelps

    Do they mean that their chips will outperform the dual-core Intel at a raw ops/s level, or do they mean that when running emulated x86-64 apps on Windows on Arm, those apps will be faster than on (older) native Intel chips?

  5. iPhoneX

    Let's not forget Apple's early involvement in ARM with the Newton, and how their own stellar semiconductor lab might be building something better than anyone else can create.

  6. skane2600

    Fortunately as a privately-owned company, ARM didn't have to include a forward-looking disclaimer although they are making forward-looking statements. Of course one can make a proof of concept 7nm or 5nm chip and still not make a high-yield device available to the market. Note, that I'm not claiming they can't do what they are claiming, but we know from the delays Intel has suffered that things don't always turn out how they are planned.

  7. waethorn

    So long as they completely drop predictive processes in the processor, then they can break Intel's monopoly easily.


    If they keep doing this fake performance boost stuff where *DAILY* security holes are found, they'll continue to falter like Intel. OEM's are starting to disable Intel SGX now by default because of them. SGX, if you didn't already know, is short for "Software Guard Extensions" - an Intel security feature. Yet it's actually making systems LESS secure having it enabled.

  8. jimchamplin

    If they can pull this off and go toe-to-toe with Intel, I would love to see what they can do with a true desktop ARM CPU. One without the thermal and power constraints of a mobile part.


    I.E. If they can make an ARM-powered notebook that meets a contemporary i5, they could really smoke Intel on the workstation and gaming PC front.


    Their problem so far is that their SMP has been lackluster.

  9. markld

    ARM is only going to make the desktop and all other devices better, even if they are not ARM. Intel needs competion

Leave a Reply