An Intel-Powered Mobile Future

Img51493_medium

via www.esaitech.com

Back in 2006, Intel had mostly lost the interest of the PC hardware enthusiast. Despite the Pentium 4’s ability to reach astronomical clockspeeds, for nearly three years their consumer line of desktop CPUs was consistently outperformed by competing products designed by AMD. The Pentium 4 was based on Intel’s NetBurst architecture, designed to push clockspeeds and ideally performance, continuing Intel’s success from previous iterations of the Pentium. At first this worked out, but with time physical limitations began to emerge and were exploited by AMD. The last Pentium 4s were clocked at 3.8Ghz, but were known for being hot, inefficient chips that couldn’t outperform AMD Athlon CPUs running at much lower frequencies. Intel knew this and was deeply discounting Pentium 4s to make them more desirable to consumers. That wasn’t their only strategy, however.

In a bold move, Intel scrapped the entire NetBurst architecture and focused their attention on the Pentium M, a chip designed for laptops that was a descendant of the P6 architecture (used from the Pentium Pro through the Pentium 3). They repackaged the Pentium M architecture as the Core Duo and Core Solo released in January 2006, 32-bit mobile CPUs that powered the first Intel Macs. Just seven months later they released the 64-bit Core 2 Duo, a major revision of that architecture running at modest clockspeeds, but clearly outperforming the Pentium 4 and AMD’s best.

Since that time, Intel has used its size, engineering talent, and manufacturing advantage to not just sustain but increase its performance and efficiency lead in the server, desktop, and laptop markets. I believe that over the next couple of years Intel will use its resources to create a similar story in the mobile space. Given their current position in the market (nearly non-existent), there’s good reason to be skeptical of this claim. However, Intel has made it clear that it’s aiming its guns straight at the phone/tablet markets, and I think they have a good chance to succeed. Here’s why.

Intel has the advantage
It almost seems like Intel is cheating when it comes to making CPUs. For many years the company has made chips with smaller parts than anyone else. This is important because those smaller transistors use less current at the same frequency, allowing you to either do more for the same overall power usage or use less power to do the same amount of work as your competitors. Given a similar architecture, Intel’s chips can be faster, more power efficient, or a little bit of both.

To give you an idea of Intel’s manufacturing lead, the company released CPUs based on its 32nm manufacturing process in March 2010 -- almost two years later in late 2011, a small trickle of 32/28nm chips manufactured by Intel’s competitors began to enter the market, with substantial volumes not available until the second quarter of 2012. Intel, meanwhile, released its first 22nm chips in April 2012.

Intel has other advantages as well, from their relationship with Apple to an apparent edge in design and execution, releasing a new CPU architecture revision every year since the Core 2 Duo. So far, Intel has never used any of these advantages for its current phone and tablet product, the Atom.

Intel_roadmap_560_large_verge_medium_landscape_medium

via cdn2.sbnation.com

The placeholder
In April 2008, Intel released the first Atom CPUs on their five month old 45nm process. Originally designed to power a new class of devices Intel was calling MIDs (mobile internet devices), the Atom eventually became the standard netbook CPU choice. A 32nm Atom was introduced in late 2011, almost two years after Intel began shipping 32nm products. Despite different packaging, GPU pairings, and other subtle differences, the Atom architecture remains essentially identical today, nearly five years after its introduction. Compare this to the 12 month release cadence of Intel’s Core line and you can see how little effort Intel has put into the Atom line.

Yet, today the Atom is Intel’s cellphone and tablet product. And guess what? Even with a woefully deficient GPU it’s actually not so bad! Not exactly removing socks from people’s feet, but for a five year old design built from Intel’s 2010 manufacturing technology, it’s surprising what Intel has been able to do while not really doing anything. Anandtech compared Qualcomm’s 28nm Krait based Snapdragon to the 32nm Atom and found that the Atom outperformed the Krait CPU while being competitive with power consumption. Another Anandtech article showed that the Atom was more power efficient and much faster than a Tegra 3.

The Atom is what happens when Intel phones it in. I’m pretty sure Intel was just using the Atom here to get their name in the smartphone space, and get a little momentum around the idea of Android running on an x86 CPU. The 32nm Atom is just a placeholder as Intel readies its next big moves.

The power difference
Before we get to that part, I think it’s important to understand some of the differences between phone/tablet chips and the ones Intel has been making for notebooks, desktops, and servers.

The most obvious and dominating factor is how much power these chips consume. The industry standard measurement for power consumption is TDP (thermal design power), which represents the most power a CPU will draw at a given time. The figure is used so that devices can be designed to dissipate heat from a chip in a worst-case scenario. For a while the TDP for Intel notebook CPUs was between 35W and 45W, with desktop TDPs going well over 100W. Recently Intel has produced lower voltage versions of their notebook CPUs with TDPs of 17W, used in the Macbook Air and the Surface Pro.

The Atom line ranges from around 2W to 10W, with the 2W end representing the latest 32nm parts. Unfortunately it’s more difficult to provide precise TDP figures for ARM chips, but thanks once again to Anandtech we can use some general numbers -- tablet SoC power consumption tends to be below 5W, and smartphone below 2W. It’s easy to see that even though Intel is dipping its toe into the ARM power consumption realm with the Atom, that even their most efficient Core chips are in a different league of power consumption.

The second difference to consider is performance. This is extremely difficult to quantify because there are very few comparisons between something like a Tegra 3 and a Core i7, and even with what’s available, there are far too many variables to really make any fair comparisons. I’m going to ignore these problems and attempt to illustrate the performance differences with chip complexity and a single, flawed datapoint.

With Atom and Core, Intel really has two different processor architectures that it sells. The Atom microarchitecture probably has more in common with the first Pentiums than even the older Core 2 Duo. Most Core i7 CPUs have one to two billion transistors, compared to the Atom and ARM SoCs which hover around 100-200 million.depending on the configuration (Apple being a probable exception). This additional complexity is not just more cores, but more complex cores that can do more in a CPU clock cycle than simpler cores can. That flawed datapoint I mentioned that I will use to illustrate this is Sunspider, which measures javascript performance. The current top performing mobile SoC-browser pairings (including Atom) score around 1000ms in the benchmark, but in more demanding desktop browsers the high end Core i7s score around 130-150ms. There are a lot of problems with just comparing those two numbers, but I’m attempting to illustrate that the Core series from Intel represents a completely different class of performance than what you typically find in phones and tablets these days. Even though a Tegra 3 (around 1700 ms in Sunspider) is a quad core CPU running at 1.5Ghz, it’s operating in a completely different performance league than a dual core Core i7 at 1.7Ghz (around 200ms in Sunspider).

With that somewhat tedious explanation out of the way, we can look at what Intel is going to do from here.

2013, the setup
Intel knows the situation. They almost certainly know that everyone predicts that mainstream computing is shifting to our mobile devices, and they also know that despite some surprisingly adequate Atom-powered products that they don’t have much consumer or OEM interest for their phone/tablet products. So, Intel finds their business moving away from them, and they’re something of an underdog once again.

The Intel machine likely swung into motion to tackle these problems many many months ago, but the fruits of this effort are only going to begin hitting the market this year. Intel is using their two parallel product lines (Core and Atom) to seriously address the mobile market. But first they need to stall just a little bit longer.

Vjb_0207_large_large_large_verge_medium_landscape_medium

via cdn0.sbnation.com


Early this year Intel announced that they would be releasing a 7W version of their Ivy Bridge CPU, the 22nm CPU Intel released last year. First, this was a little misleading since the 7W figure is not the TDP for this new chip but a new figure Intel is calling SDP (the actual TDP will be 10W). While this is certainly a more compelling product for something like a Surface Pro tablet than the 17W Ivy Bridge, it still doesn’t bring the Core line down to a more tolerable tablet size, like the iPad or Surface RT. I believe this is a bit like the current 32nm Atom, in that it is just Intel getting a product out there to drum up interest in x86 tablets, and give OEMs something more reasonable with which to build.

Dsc_0352-875_medium

via assets.sbnation.com


Now enough with the stalling, it’s time for the good stuff. Later this year, Intel will release the first new Atom architecture since 2008, and it will be manufactured on Intel’s 22nm process that it introduced in 2012. While we don’t have a lot of information on what to expect with the new Atom, from what we do know it should be a much more modern architecture. It also seems clear that Intel is finally taking the Atom more seriously.

In 2013 the new Atom will be facing off against new ARM A15 SoCs (first introduced with the Chromebook and Nexus 10 last year), as well as a new revision of Qualcomm’s Krait and likely Apple’s A7. While the A15 does outperform the current Atom by a decent margin, I think chances are good that Intel can produce at least similar performance, if not better with their next Atom. We know that the new Atom will finally execute instructions out-of-order (something ARM chips have done since the A9), and will adopt Intel’s 64-bit version of their x86 instruction set. These are major enhancements, but really only give use a peek at how the new Atom will perform. That said, given how the 2008 designed Atom running on Intel’s almost three year old process stacks up against the bleeding edge A15, I’m thinking Intel would really have to blow it for a modern 22nm Atom to not compete very well.

Unfortunately for A15 class ARM SoCs, Atom is not the only thing they have to worry about.

Some pun using the word Haswell
While Intel has certainly made numerous power efficiency improvements over the years, it’s arguable that Intel’s Haswell is their first CPU that demonstrates their commitment to putting their most robust designs into handheld devices. To be released this year on their 22nm process, Haswell promises to bring extremely aggressive idle states to a Core class CPU. Intel is promising a twenty times improvement in idle power consumption. It’s also expected that a low voltage Haswell part might be available with a 7 or 8W TDP. That’s huge.

If you recall, I mentioned that tablet SoCs tend to operate below 5W. That figure doesn’t tell the whole story, and likely doesn’t represent what would be the TDP of an ARM SoC in a tablet. Anandtech did some power consumption testing on the Samsung A15 SoC in the Nexus 10 and determined that while power was generally throttled down to 4W under load, that power would peak up to 7W. This 7W peak is more akin to a TDP than the 4W figure, which you might say is analogous to the SDP Intel came up with. If Intel really can release a 7 or 8W TDP Haswell in 2013, they might just have a Core product that can compete directly against the upper-end of the ARM range.

There are some caveats here. First we don’t know for sure if Haswell can really hit those numbers. Second, Haswell likely isn’t going to run as a fanless CPU unlike these ARM SoCs, so there will still be differences in the form factor even if power consumption is identical.

To recap, in 2013 Intel will for the first time be truly competing in the tablet and smartphone markets. They will have a brand-new Atom that will likely equal or exceed A15 class chips in performance and/or power consumption, and will have a low-power Core chip that could likely fit in something the size of an iPad 4. It seems with Atom powered phones, and Atom/Core powered tablets, Intel has a good chance to produce products that are more compelling than what ARM SoC vendors are bringing to the table. I submit to you that this is just the teaser for 2014, when things get really interesting.

14 in 14
Two big things are scheduled to happen in 2014. Intel will release their follow-up to Haswell known as Broadwell on their new 14nm process, and Intel will release an Atom based on their 14nm process. The new Atom is significant because it represents truly putting all of Intel’s muscle behind the Atom. For the first time, it will be competing against ARM smartphone designs with Intel’s full manufacturing advantage. It’s true success hinges on what the new Atom architecture in 2013 looks like, but with a 14nm Atom slated for 2014, it’s abundantly clear that Intel is taking the Atom line very seriously.

You can think of Broadwell as essentially Haswell at 14nm. Intel’s CPU releases generally alternate yearly between subtle tweaks that accompany a new, smaller manufacturing process, and a brand new design using last year’s process. Broadwell will be one of those subtle tweaks on a new process. What’s interesting to consider is a Haswell chip that requires even less power to do the same work. It’s quite possible that this CPU could work in fanless designs and could easily fit in slim tablets. This is likely where Intel produces a killer tablet chip. Whereas Haswell tablets might be slightly heftier than ARM counterparts, Broadwell will likely bring ergonomics nearly identical to an ARM tablet while bringing truly laptop-class performance.

What’s perhaps even more interesting is in 2016 when Intel moves to the next smaller manufacturing node. We might see Core chips in smartphones at that point, pushing Atom down to even smaller devices.

Some loose ends
I’m excited about this future, one with desktop-class performance in my tiny phones and tablets. However, Intel has a few big problems to sort out first.

Core series chips are expensive. Among those 7W Ivy Bridge CPUs that will be released this year, the cheapest is $150. One is $362. ARM and Atom SoCs typically cost between $30-50, and usually on the low end of that. Some of this has to do with the size of the chips -- the bigger a chip is, the fewer of them you can put on a wafer, and the more expensive that chip is to produce. However, with the massive size of Apple’s SoCs (in particular the A5X and A6X), this factor alone shouldn’t stop Intel from being price competitive in this market. The real problem is obvious: Intel doesn’t want to devalue their laptop and desktop CPUs by offering essentially the same chips at a fraction of the cost.

In my opinion, Intel just needs to dive in here and take the risk. Intel Core tablets aren’t going to blow minds if you have to pay a $200 premium to get them. Intel’s business is all about the high margin enterprise and enthusiast consumers (much more the former than the latter). The other CPUs they produce are all about manufacturing in high volume which affords them the ability to execute well in the high margin markets. They need to transition their volume manufacturing to a more mobile class of devices. Clearly that’s what they want to do based on their roadmap, but part of this transition will be pricing their products for this new market. Atom makes the challenge a little easier in the short term, but at some point they are going to have to move to a new pricing model for their Core consumer products.

The second big problem they face is what used to be their bread and butter -- x86. With every major mobile OS and app built to run with the ARM instruction set, Intel finds themselves with their only major market disadvantage. A year ago, this one looked a bit more daunting, but with the decent user experience of the RAZR i, I think Intel has a good shot of at least shifting Android to a solid ARM+x86 future. Windows 8 is obviously an asset here (full disclosure: I’m a former Microsoft employee), moving toward the tablet world with its UI, but bringing all the richness of the x86 Windows ecosystem with it. Still, this is an unproven part of the plan, and there are more platforms in the market than just Android and Windows. Intel needs to pave their way into as many phone and tablet designs, as soon as possible with the Atom so that when the Core series is ready they can just offer that x86 solution instead.

While not strictly a problem, there is another unproven element to this story -- the GPU. Increasingly this is becoming one of the most important parts of mobile computing. Apple seems to emphasize this in all their SoCs, and other SoC makers are starting to place a similar emphasis on this part of their design. Despite selling integrated GPUs for years, Intel is not known for producing excellent products in this area. It wasn’t until the Sandy Bridge CPUs released in 2011 that Intel had a decent product, but even Ivy Bridge doesn’t come close to discrete graphics performance (look at the 13" retina MacBook Pro).

Intel has made some big promises about GPU performance in both Haswell and Broadwell, and despite opting to use Imageon GPUs in their recent Atom SoCs, they’re planning to use their own GPUs in the next generation Atom parts coming out this year. They look poised to deliver here, but this is somewhat unproven territory for them.

Lastly, ARM isn’t resting with A15. At the end of next year, or early 2015 designs based on ARM’s A57 will start hitting the market, which promises a 30-50% performance improvement over A15, along with support for ARM’s new 64-bit instruction set, ARMv8. It’s likely Apple and Qualcomm will have a more competitive Swift and Krait at that time as well. If one or more of these designs is really exceptional, and Intel fumbles its execution, it’s possible this could play out differently. That said, it still appears Intel will have the edge here.

Ok, wrap it up
As a hardware nerd, I’m completely thrilled to see Intel mobilize so heavily (accidentally awesome pun) toward a new goal. Historically whenever Intel is most challenged, that’s when they release their most amazing products. ARM and derivative designs from Qualcomm and Apple have had this new mobile market essentially to themselves for nearly six years now, albeit with fierce competition among them. Intel justifiably feels that their business is threatened, and over the next 18 months we will see their response in the market. I think there’s a good chance that the fruits of this will be an even more accelerated advancement in mobile devices, along with a more diverse set of devices and experiences. Now we just need to see if Intel can really pull it off.