2014 is set to be a year of tremendous change for both Intel and AMD. Both companies are responding to rapidly shifting market dynamics as the computing market continues its greatest product transition since the PC debuted nearly forty years ago. The two CPU manufacturers are attacking this shift from different directions and with different product strategies – we’ll have a look at both here, starting with Intel first.
Intel goes for tablets, 2-in-1s
For 2014, Intel is focusing on what it calls 2-in-1s – systems with both docks and tablet components that snap together or pull apart. That might seem like a gimmick, but the way in which these designs are presented to the mass market reflects a fundamental change of thinking. In the past, 2-in-1 systems have been tablets first and foremost with overpriced docks that offer minimum functionality, yet cost £100-odd for a USB port and a keyboard. This has changed. Modern x86 2-in-1s integrate the dock as standard and don’t slap on an enormous price increase for doing so.
When Microsoft launched Windows 8 in late 2012, ARM tablets were supposed to take the low end with Windows RT, while x86 tablets based on Clover Trail occupied the $500 (£300) to $800 (£500) space, and Ivy Bridge-based tablets took the $800 (£500) plus market. This failed rather impressively. ARM-based Windows RT tablets didn’t establish a buffer for Intel’s Clover Trail; they failed to sell. The Clover Trail-equipped systems, meanwhile, offered terrible performance compared to existing laptops at the same price points.
Now, that’s changing. Intel has begun aggressively backing Bay Trail and Haswell-based systems across multiple operating systems and at price points that compete directly against ARM-powered Android tablets. Intel isn’t just focusing on Windows, either – we’ve already seen Haswell-based Chromebooks, and we’ll see more support for Android tablets arriving in 2014.
The change has been driven by the realisation that Atom cannot continue to be treated as a threat to Intel’s Core business. While Haswell made impressive power consumption gains, Intel still needs a chip that can sell into fanless systems and that serves as a lower-end, lower-cost buffer for its most expensive parts. Strong competition from ARM means that Chipzilla no longer has the luxury of creating mediocre-performing hardware year after year and proclaiming it a winning strategy.
2014 is shaping up to be the year that Bay Trail transforms Intel’s x86 business in both servers, mobility, and tablets. The only way to pull off such a shift, however, is for Intel to preserve higher margins at the high-end by offering an even better chip.
The Broadwell family
Next year, Intel will offer multiple Broadwell product families, including a rumoured Broadwell-U (Ultrabooks) and Broadwell-Y (tablets). The tablet chips will reportedly target fanless operation, and offer a TDP of 4.5W and an SDP of 2.8W. As for how effectively this will compete with chips from ARM, it’s better to focus on the “fanless operation” data than the TDP figures.
Pushing Core-derived chips all the way into fanless chassis is a major achievement. One of the problems with Microsoft’s Surface family, for example, is that the Intel-derived parts are slightly thicker and noisier due to the need for a cooling fan. If Intel can offer these improvements without the corresponding increase in weight that we’ve had to put up with until now, then Broadwell could catch on with professional customers who don’t want to skimp on performance, weight, or battery life.
AMD’s multi-pronged assault
The major difference between Intel and AMD is that Intel is shifting its Atom strategy to fend off ARM in data centres while simultaneously attacking in tablets. AMD is launching (or has just launched) entire new product families to hopefully seize market share.
The console question
There’s no way to predict exactly how much money AMD will make off consoles in 2014. We know that both Microsoft and Sony claimed more than two million sales apiece within the first few weeks of availability, but not much else. If we assume total Christmas sales of six million systems between both manufacturers and use the estimated $100 (£60) to $110 (£65) cost of the AMD chip combined with AMD’s comments on its operating margin for the semi-custom business (during the Q3 conference call, AMD CEO Rory Read said that margin was in the “mid-teens”), then we can assume AMD makes between $14 (£8.50) to $16 (£9.70) per console sale. That works out to a pre-tax profit of some $80 million ($49 million) to $100 million (£60 million) per quarter.
For AMD, that’s a healthy chunk of profit, and it could definitely help pad the company’s income during a tumultuous transition period, if console sales remain strong overall.
AMD’s ARM server play
AMD’s first Cortex-A57 chip, codenamed Hierofalcon, is supposed to debut next year at 15W to 30W with up to eight CPU cores and support for 10Gb Ethernet. The stats on the core and its associated TDP indicate that AMD wants to hit Intel in terms of raw performance, not just a hyper-optimised power play.
With Calxeda gone, it’s possible that AMD will step into a role as one of the premiere ARM embedded vendors in 2014, provided that demand materialises for these types of solutions in the first place. AMD’s ownership of SeaMicro gives it an advantage in one respect – it can build the solutions it wants to see brought to market, rather than attempting to cajole manufacturers into creating them.
What’s interesting about Hierofalcon is that it supplants Kabini-based Opterons in the server market altogether, with AMD earmarking its low-power x86 chips for a spot in the 5W to 25W low-power embedded market, relying on ARM for its enterprise server clusters, where Intel intends to compete with Bay Trail.
HSA finally comes to market
Finally, there’s Kaveri and AMD’s first HSA-enabled chip to hit this space after literally years of work. Kaveri-based chips will launch for both consumer and server markets in fairly short order, replacing the existing quad-core Piledriver chips and (hopefully) offering a mixture of higher single-threaded performance and the option to use the GPU as an accelerator in a variety of tasks.
Ultimately, we expect Kaveri’s HSA capability to have only a very modest impact on AMD’s APU sales. HSA faces the same problem as any new standard; no matter how awesome eventual performance might be, no software exists in the present day to take advantage of the function. Oracle will add HSA support with Java 8 this year and project Sumatra, but native support for HSA execution won’t appear until Java 9, currently expected in 2015.
It’s going to be harder for AMD to build support for HSA in the x86 world, given that it commands only a minority of that market. Some ARM vendors, like Qualcomm, may also adopt HSA on future products, but being first out of the gate often means that early support is never as useful as early adopters would like.
HSA is best understood as a long-term play that AMD hopes will become a vital standard of heterogeneous system design, much as x86-64 shaped more than a decade of CPU advances. Kaveri sales will be driven by more prosaic improvements on the GPU and CPU front, and possibly by Mantle’s performance.
Seeking new footing on treacherous sands
Intel and AMD are both trying to reposition themselves to better respond to and anticipate growth in the computing markets of the future. Of the two, AMD has gone further in its quest to reinvent itself, but it’s also in the more precarious position. AMD has bet on consoles, on HSA, on ARM, and to a lesser extent, on Mantle. If these offensive manoeuvres don’t win it sustained market share growth and higher revenue, it’s going to be in serious trouble.
Intel, meanwhile, has the luxury of playing a more measured defensive game. While it clearly takes ARM’s threat in the data centre seriously and still wants to be an active player in mobile markets at some point in the future, the failure of Windows RT and the lack of any major data centre opponents gives Chipzilla time to consider its course.
What’s striking about 2014 is the degree to which AMD and Intel are looking past each other for opportunities in areas. AMD has Mantle and its console business, while Intel is working to support no fewer than five operating systems (Android, OS X, Chrome, Linux, and Windows). All of AMD’s discussions of tablets remain Windows-focused, which makes sense given the company’s more limited resources.
The x86 market may not face the ARM challenges from Windows RT that we once thought it might, but both Intel and AMD are pursuing new growth strategies in the face of the ARM threat. At the end of this year, we’ll know a great deal more about which of those strategies is likely to succeed.