Shortly after CEO Paul Otellini announced his retirement, a number of roadmap and product strategy leaks occurred. Far be it for me to determine whether this is an intentional move intended to silence critics who claim Intel (INTC) has no mobile strategy or pure coincidence, a lot of good stuff has made its way onto the internet that investors - especially those who don't scour the web for every last scrap of information - be made acutely aware of where Intel is going. The buy-side analysts have likely digested this information and are planning their investments accordingly, so the retail investor needs to be up to speed, too.
Bay Trail - Coming To Tablets Near You In Early 2014
Here's the first juicy slide:
In this slide, we can clearly see that the next generation "Bay Trail-T" (presumably the "-T" is for "Tablets") will be hitting tablets in early 2014, roughly 1 year from the Cover Trail (Atom Z2760 in marketing-speak) launch.
The Atom Z2760 has scored a number of design wins including:
Clearly, Intel has a viable tablet SoC (although many media pundits would like to tell you that Intel has no products suitable for mainstream tablets). However, things get really interesting with "Bay Trail."
Bay Trail - Performance, Power and Features All Get Supercharged
While the previous slide simply told us timing, it told us nothing about the actual chip and platform. Luckily, the leakers were kind enough to provide investors with the following slide:
This is where things get good. The bullet-point features are the following:
- CPU performance-per-watt goes way up: at the same performance level as this year's "Clover Trail," power consumption gets cut in half. At the same power, we are to see a whopping 50-60% performance improvement. Considering that the current "Clover Trail" 2-core/4-thread CPU is 30-50% faster than Nvidia's (NVDA) Tegra 3 which packs 4 of ARM's (ARMH) Cortex A9 cores, this is no small feat.
- Memory Bandwidth Doubles: A disadvantage that Intel's "Clover Trail" has against next-generation ARM Cortex A15 SoCs such as the Samsung Exynos 5250 and the Apple (AAPL) A6X is that theoretical peak bandwidth for "Clover Trail" lags (although Atom's excellent caches and efficient memory controller mitigate this theoretical disadvantage quite drastically). Memory bandwidth determines how fast the CPU and GPU can be fed, and any bottleneck easing here is going to lead to substantial gains, especially as the new architecture is likely to be significantly more powerful, thus requiring more data per cycle.
- Graphics Performance Improves Drastically: Intel's weak spot in its released mobile processors (Atom Z2460 for phones, Atom Z2760 for tablets) is in its graphics. Currently, Intel licenses graphics cores from Imagination Technologies, but it seems that in the next generation Intel will be leveraging its own in-house graphics tech. This is important for a number of reasons: first, gross margins improve as licensing fees to Imagination disappear. Next, Intel will leverage its graphics R&D work (hardware and software) for its mainstream notebook/desktop chips, which should provide a better return on that investment (and minimal OpEx increase from moving to in-house tech for tablet chips). Finally, with better graphics (that are at least 3x faster according to the slide than the current gen), higher resolution displays can be supported ("Clover Trail" supports 1080p but "Bay Trail" lets OEMs build 2560x1600 display devices). Credible competitors to the iPad 4 (and beyond) can be finally be built with "Bay Trail"
In short, "Clover Trail" is competitive today at the 2W level, but it is not a clear slam-dunk winner compared to the Apple A6 and Samsung Exynos 5250 based on the Cortex A15. Competition from the Nvidia Tegra 4 (based on the Cortex A15 and Nvidia's own "Kepler" graphics technology) will also be quite fierce in the middle of 2013. Intel's first shot at establishing leadership lies with "Silvermont" and "Bay Trail."
So the question, then, is, "well, OK, so why is Intel even going to have an advantage going forward?"
The answer relies on two fundamental facts:
- Intel's experience in designing higher performance processors
- Intel's process technology lead (22nm tri-gate versus 28nm/20nm planar)
Experience Goes A Long Way: As tablets scale up to become more powerful, the processors themselves will fundamentally start to resemble the high end processors from 5-10 years ago. While the ARM guys are coming at the problem from the perspective of low power, low performance parts that need to scale up (which relies essentially on die-shrinks and power gating with known-quantity microarchitectural principles), Intel is coming at the problem from the other end: it has the know-how to build very fast and power-efficient processors at the high end, but needs to scale down to lower wattages.
One requires inventing fundamentally new techniques, the other requires scaling down yesterday's performance into today's low-power envelopes. The former is likely much more difficult than the latter, especially on a budget, and on last-generation process technology.
The Process Technology Lead: Having a process technology lead doesn't just mean that you can produce the same things as your competitors but cheaper, but it fundamentally enables more features at the same cost level. When designing a micro-processor, the architects need to shrewdly pick and choose which features can fit the die size and power-budget constraints (among other things, including difficulty of implementation). Things that are too power-hungry or costly at 32nm/28nm are much more feasible at 22nm tri-gate and even more so at 14nm.
What Does This Mean?
In a situation where engineers are restricted to a certain number of watts and a certain die size, the ones with the smaller, faster, less-power hungry transistors at their disposal will naturally be able to produce a faster, more-efficient design.
This means that Intel can out-price the competition at the same performance level while maintaining comparable gross-margins or it can offer fundamentally superior products at higher price-points.
The Achilles Heel #1 - Married To Microsoft's Windows 8
While I believe Intel is making all of the right moves on the hardware side of things, there are two major causes for concern on the part of investors. The first concern is obvious from this slide:
While the increased battery life, hugely better CPU and GPU, and DirectX 11 support seems awesome, note that the OS coverage is limited to Windows 8. That's right - no Android.
While Microsoft essentially gave Intel the finger by building Windows RT (ARM-compatible Windows), it seems that Intel is trying to get back on Microsoft's good side by limiting tablet support to Windows 8. I'm a firm believer that Windows 8 is an excellent product for both tablets and notebooks, but my view is not necessarily consistent with the market's view.
If tablets are to be thought of as simply an extension of the laptop form factor, then this decision makes perfect sense as Windows 8 tablets would, in this case, overtake Google's Android and Apple's iOS due to the sheer popularity of Windows 8 on the PC side.
But if tablets are fundamentally different, supplemental, creatures, and NOT potential netbook/low end notebook replacements, then adoption of Windows 8 tablets is still a huge unknown. This is something investors need to watch quite closely.
Achilles Heel #2 - Timing
Another big problem is the timing. While a previous roadmap from Intel noted that the next generation "Silvermont" core would hit in 2013 and its 14nm successor, "Airmont," would show up in 2014, it is clear that - at least in tablets - that's not going to happen.
"Silvermont" will appear in tablets during 2014 and "Airmont" will likely appear 1 year later. This means that Intel's 22nm tri-gate part will compete with 28nm/20nm planar Cortex A15 designs from ARM and custom ARM-compatible designs from Qualcomm (QCOM). While Intel still has a slight process advantage thanks to tri-gate (which shines in low-power applications), it will not be as pronounced as it would have been if "Bay Trail" was a mid 2013 part. There are very few details known about the "Silvermont" micro-architecture.
The unfortunate consequence is that Intel will spend the entirety of 2013 fighting modern ARM parts with its "Clover Trail" SoC. Against the Cortex A9 and the Snapdraong S4, the current Atom does fine, but the Cortex A15 and perhaps a next generation Snapdragon would seriously put a damper on the desirability of Intel's products.
Intel cannot fight with "good enough." If it is going to win the tablet wars, it needs to have the best silicon available to OEMs. Period.