Some while back, Wolfgang Grüner posted an interesting hypothesis about AMD’s future direction possibly including ARM designs, at Conceivably Tech (original article) and many others have chipped in saying this is either a terrible idea or a great idea.

So, could AMD be about to give up their crown jewels, x86, and jump ship to ARM to compete in the growing smartphone and tablet markets with dozens of other chipset manufacturers?

Well, AMD say no, but had previously been speaking in somewhat coded terms and have followed this up with several very vague comments about no doors being closed. With no permanent CEO in place, it seems unlikely that such a strategic decision could have been made in the past few months but why should (or shouldn’t) AMD consider bringing out an ARM chip in theory, if they haven’t decided this already behind closed doors?

We’ll have a look at if this is a sensible plan or whether they may be trying to buy some time: announcing new ARM products may well hurt their current product offerings and roadmap, yet delivering a whole new CPU architecture on a new instruction set takes a long time and, if they were to have only started recently, they would be some years away from hitting real products.

x86‘s past

In the past, AMD’s main strength has been its x86 licence which has allowed it to run Microsoft’s Windows and the vast library of applications available for that, letting them compete against Intel for the high ASPs, decent margins and good volumes offered in the PC world.

VIA, Cyrix, Rise, IBM, IDT, Transmeta and others who previously sold x86 processors have effectively fallen to the side leaving a virtual duopoly of Intel and AMD; even AMD looked to be on shaky ground until new investors from Saudi Arabia spun off their manufacturing division and Intel paid them $1.25 billion to settle a massive anti-trust legal case.

In recent years, ARM tried to break into the PC market, showing off tiny, low-cost netbook-competitors to little effect. Qualcomm invested large amounts in promoting their SmartBook concept and, again, failed. Why? No O/S. No apps: no Office, no IE, no iTunes, no Photoshop, no Firefox, no Flash.

It didn’t matter what chip you had inside your product: if you released something which looked like a notebook PC, people expected Windows. Linux was not ready for the unwashed masses, as Acer discovered with their first EeePC models, where the Linux ones were sent right back to the stores they were bought from.

But times have changed, with the iPad redefining what was needed from a computing device and with Microsoft announcing that future versions of Windows would run on ARM as well as x86 – the first open, dual-architecture consumer platform in the desktop Windows range. Even Apple are rumoured to be making a switch to ARM so maybe the time is right now?

Market followers?

In fast-shifting industries, being behind the cutting edge can both be a blessing and a curse. It’s frequent that the first to market has failed to make the most of their position and become dominant. But missing the boat altogether is categorically a course for disaster.

It’s not too controversial to suggest that AMD have traditionally followed Intel’s lead, with the exception of a period where their K7 and K8 chips led the way with innovations including bringing 64-bit x86 extensions, on-die memory controllers, Silicon-on-Insulator and monolithic dual core CPUs to the mainstream.

Sadly for them, they failed to translate this period of strength into a transformation of the company into a genuine leader in the x86 field, appearing uncomfortable with their new standing and misfiring on several follow-up products, even allegedly ripping up major parts of their roadmap (such as the K9) and allowing Intel to spectacularly re-capture the performance lead with their Core 2 microarchitecture.

AMD have also often been 12-24 months behind Intel with manufacturing, support for new instruction sets, features and overall performance; they have usually achieved sales through lower prices, Just Good Enough performance and fewer restrictions.

For example, while an excellent product, AMD’s competitive answer to Intel’s Atom (Fusion) only hit the market in late 2010: more than two whole years after products using the Atom first went on sale in early 2008 and after the sales of netbooks (Atom’s bread and butter) had started to decline. Before this, they focused the majority of their efforts on desktop chips just as the world was shifting to laptops.

The technology world sadly has many examples of boats that AMD managed to miss. Is mobile to be another of these?

All change

The tides of the technology world have shown major signs of shifting, with Intel announcing today their new focus on mobile, and, while traditional PC products still account for massive volumes, the momentum is certainly with mobile-derived products at the moment. Here, AMD have already missed out yet again, having sold off their ultra-low power CPU division to Raza and ATI’s mobile graphics division to Qualcomm, ironically forming one of their strongest competitors today.

It’s almost certain that CEO Dirk Meyer’s departure was not just down to delayed execution on their “Fusion” products, Atom’s competitor, but that he had failed to provide a competitor for the smartphone and tablet markets, where all the growth was. Certainly, AMD’s executives are on record as desiring a big piece of the tablet market.

Where’s the software?

While AMD’s Fusion products look to be a good fit, hardware-wise, for tablet products (and their 28nm follow-up products even more so), they now find themselves on the wrong side of the software fence: with Microsoft’s tablet strategy backfiring so badly, AMD’s x86-compatibility is now a liability in the world of mobile Operating Systems and applications, designed to run on the ARM architecture.

This has allowed competitors Qualcomm, Broadcom, Mediatek, Marvell, TI, Freescale and OEMs like Apple and Samsung to produce chips for personal computing devices (that’s with a small P and C) and fight for the business amongst each other with both Intel and AMD sidelined like never before.

Look at the rate of new ARM apps appearing on iOS and Android compared to x86-compiled ones in the world of desktop Windows and Mac OS X. Even bringing the App Store model to OS X does not appear to have delivered the same explosive growth and innovation as in mobile.

In a nutshell, unless they go ARM, AMD currently desperately need a successful mobile operating system with apps for their x86 chips to be relevant in this new world. For that reason, AMD joined MeeGo which, with Intel and Nokia on board and a very different set of values to the relatively locked down iOS and Android (for good or bad) providing strong diffentation, looked to have a decent chance of success.

Nokia’s departure from MeeGo appears to have effectively closed the door on that operating system as a competitive mobile OS which runs equally on ARM and x86. If MeeGo does indeed fail, it will hurt both Intel and AMD.

So how about Android? Well, even on Android, despite having many apps written in cross-platform Java, x86 is incompatible with the kind of apps for which high-performance CPUs would most be of benefit, which have been developed using Android’s NDK (Native Development Kit) and compiled into ARM code.

Categorically, these apps (which include many games like Angry Birds, augmented reality and other performance-dependent apps) will not run on an x86 chip unless some form of emulator is provided which would hurt the performance.

If Intel’s current strategy does work and they do manage to bring the momentum to x86 in mobile, then AMD will be OK with continuing on today’s path. But that’s by no means guaranteed and it’s a mighty risky strategy to rely so much on your main competitor. By shifting to ARM, if the x86 in mobile strategy fails, AMD would at the very least have a head start on their main competitor if Intel also were forced to shift.

How would / should AMD differentiate if they go ARM?

Let’s make no qualms, there is little overlap between x86 and ARM in CPU performance at the present moment. The highest performance ARM CPU core at the moment is the Cortex-A9 with Qualcomm’s Scorpion not far behind. The next generation scorpion core will undoubtedly reclaim the lead and ARM’s Cortex-A15 will respond to that. But even the fastest quad core Cortex-A9 will be nowhere near Intel or AMD’s finest.

There are valid reasons for this, but the x86 competitors currently occupy a higher space in terms of raw performance. AMD and nVidia also hold a staggering advantage in graphics performance which is bound to the x86 world at present. Good benchmarks to compare these two worlds don’t even exist yet, the chasm is still so large.

Graphics chips are also now able to handle more general purpose code through APIs such as OpenCL. This reduces the impact of the CPU ISA and makes the change even easier.

Naturally, therefore, the obvious space for AMD to occupy would be above the existing players. Want the ultimate in low power? An AMD ARM chip may not be the right choice. But, if you can cope with slightly higher power consumption (like an ultra-thin laptop) then you are not so restricted and a 5-10W ARM chip with class-leading CPU and graphics performance may be perfectly acceptable. They have working relationships with every major OEM and plenty of manufacturing capacity and experience.

Pick a spot at the top and they can choose their price and cream off the profits from the industry, leaving others to fight with the low cost base companies like MediaTek.

But a System-on-Chip is not so much a musical instrument as an orchestra: it’s not all just about the CPU and GPU, but having a balanced ensemble of components with no obvious weak link such as an out-of-tune trumpeter screeching over the otherwise beautiful sound. In the x86 world, it’s not just the CPU that takes more power – it’s almost every single component. There are also many unknowns for AMD moving into this field – image processors, DSPs, on-chip power management, security… Clearly, there is a lot of work to be done to reach the balance and create a desirable SOC.

AMD are on the right lines with their Fusion products which are very comparable in silicon size with the latest ARM products from nVidia, Samsung, Qualcomm and ARM, but there’s still a big gap in active power consumption and an even bigger one in idle / standby power.

AMD’s options

For AMD to abandon x86 for the PC market would be foolish, unthinkable; that is certain. It may be a good few years before the mobile architectures start to seriously impact the PC world.

Like it or not, however, the SOCs are coming for their main market, at a lower price and with much more software. If they choose not to respond with an ARM product, AMD risk choosing not to take part in the future.

Taking on an ARM architecture licence and designing a higher performance core to combine with their leading graphics architectures would be no different from what they do today: taking an x86 licence from Intel and trying to design a higher performance core. It would, however, be lower cost, easier to win and come without any kind of onerous licensing terms.

This is a classic situation where hedging your bets is the best strategy by far: at present AMD’s market share in tablets and mobile phones is 0%. The only way would be up.