Without Meyer, what will AMD do next?
ARM Micro Devices
Analysis Things seemed poised to turn around for AMD in 2011. But the abrupt departure of CEO Dirk Meyer on Monday afternoon – at the exact same time that rivals Intel and Nvidia ceased their hostilities and a week after Nvidia jumped into the processor racket – indicates that AMD's board of directors sees challenges that aren't obvious to outsiders.
And to at least one ex-insider who is now looking for a job.
The chatter is that Meyer was shown the door at AMD because the chip maker had missed the boat on smartphones and tablets, markets that are growing a lot faster (in terms of units) than traditional CPUs and GPUs. As Nvidia president and CEO Jen-Hsun Huan said at the launch of his company's Tegra 2 system-on-a-chip last week at the Consumer Electronics Show, for a lot of people, their smartphone is "their most personal computer."
As El Reg previously reported, Nvidia has licensed the Cortex-A15 processor design from ARM Holdings – the company that controls the ARM RISC chip that dominates the smartphone and handheld market – and is set to announce its own multicore ARM processor, code-named Project Denver, in 2013 alongside its Maxwell family of GPU chips.
The landscape of the computing market is changing fast, and some of that shaking and quaking is due to AMD itself. AMD bought ATI Technologies in July 2006 for $5.4bn to get its hands on a GPU and chipset businesses that would help it become a platform player like rival Intel. And in March 2009, AMD abandoned its roots and chucked its chip fabs onto the shoulders of GlobalFoundries, along with rich Abu Dhabi backers who fancy controlling a business not based on oil. (AMD founder Jerry Sanders once famously observed that "real men have fabs," snarling at rivals who used third parties to bake their chips.)
Under Dirk Meyer, AMD survived the economic downturn (albeit not unscathed), dealt with product delays and changing roadmaps for standalone processors and converged CPU-GPU hybrids, and suffered through a massive platform upgrade last year with the Opteron 4100 and 6100. AMD even faced down a resurgent Intel's Nehalem family of Core and Xeon chips in 2009 and Westmere and Nehalem-EX desktop and server processors in 2010.
AMD's Fusion line of PC chips is now ready to face Intel's Sandy Bridge 2nd Generation Cores in 2011, and their Opterons arguably beat Intel's chips in terms of price/performance and performance/watt. GlobalFoundries is apparently planning to double its spending on chip-making factories and equipment in 2011, hitting $5.7bn. Oh, and AMD buried the legal hatchet with Intel back in November 2009, raking in $1.25bn and curbing some of Intel's bad behavior.
Shouldn't Meyer have been secure in his job after that hard slog over the past few years? Apparently not.
Rather than those positive developments, AMD's board focused on what Meyer didn't get right. For example, AMD chips are not the CPUs in game consoles. And worse yet, AMD chips are not in cell phones, smartphones, or tablets. AMD has somewhere between 11 and 12 per cent of global microprocessor revenues, compared to Intel's roughly 80 per cent, and it peaked in the server racket a few years back when the Opterons were so much better than their Intel Xeon competition.
The problem is that while AMD was busy cleaning up its books, going fabless, and getting its processor roadmap back in order, a slew of other products - netbooks, tablets, ereaders, and truly smart phones - changed the market. These areas are growing while PCs and servers are losing steam. To be sure, there are hundreds of millions of PC and server chips being sold each year, and that will be true for as far as any of us can see. But today there are billions of other chips being consumed, and AMD can ill afford to ignore that fact.
AMD has the Geode LX low-powered chip, which it bought from National Semiconductor in 2003, and the company could have long since created an x64 alternative to the Athlon, Turion, or Opteron for devices such as tablets. If these Geode chips were inadequate, AMD could have partnered with or acquired VIA Technologies, another maker of low-powered x64 chips, and created something that could have taken on Intel's Atom and various ARM chips for small computing devices.
But AMD – or Meyer, it seems – apparently thought the company was on the right track in the embedded processor space with its low-powered Opterons for hyperscale server clusters and the combination of the Athlon II X2 embedded processor and the Radeon HD 6700M for high-end devices such as the Surface 2.0 touch desktop from Samsung and Microsoft, which debuted at CES 2011.
AMD just released its Fusion C-Series low-power APUs (accelerated processing units), and its Fusion G-Series embedded APUs are soon to hit the streets. Both are based on the company's new low-power Bobcat CPU cores – AMD's first new x86 core since 2003 – and both feature an integrated Radeon-based GPU core on the same silicon. While these CPU-GPU APUs are impressive, it's unlikely you'll see them a smartphone – perhaps a tablet, but we'll have to wait and see.
To play the CPU-GPU limbo game, AMD could get under the power consumption broomstick in a much simpler way. The most obvious thing would be to do the same math that Nvidia did many years ago, see that at least some of the "computing" market was shifting to ARM processors, and license the Cortex designs from ARM Holdings. That way, AMD could come up with something new and interesting, like Nvidia is trying to do with the Denver multicore ARM chips for servers and their related Maxwell GPUs.
But the ARM market is rapidly getting crowded, with Calxeda and Nvidia now chasing servers, Nvidia chasing PCs, and a slew of companies – including Apple, Nvidia, Marvell, Qualcomm, Texas Instruments, and Samsung – are making ARM-derived chips for netbooks and smartphones.
For the PC and server markets, ARM chips won't be fully suitable until they have higher clock speeds, more cores, and more memory and I/O capacity. So it's not too late for the new CEO at AMD to jump into the ARM fray, or to acquire someone who is already there. Considering that Marvell now has a market capitalization of $13.2bn and Nvidia is at just under $12b, such acquisitions seem unlikely. And it's hard to imagine that Samsung, Texas Instruments, or Apple would want to buy AMD – although an Apple acquisition of AMD and a subsequent push into ARM chips would certainly be an interesting development.
An ARM license would not only be the cheapest alternative for AMD, but perhaps the only alternative – short of merging with Intel. (Wouldn't that be funny, watching Intel argue with US and EU antitrust authorities that the relevant market includes ARM devices, and that it doesn't have a monopoly on microprocessors?) An ARM license would put AMD in competition with lots of aggressive chip makers that peddle high-volume, low-margin chips, while at the same time Intel and AMD would fight the converged CPU-GPU wars on the desktop and in the laptop.
It's hard to see what AMD's next CEO will do – but it's pretty clear that it won't be an easy job. ®
Sponsored: Fast data protection ROI?