July 29, 2013, 1:55 PM — Most companies don't get a second act in tech; AMD is on its third. You have to admire this company for surviving 30 years of battles with Intel and tripping over its own two feet a few times in the process.
The first act featured AMD as a memory maker, then a x86 licensee and more or less an also-ran that lagged significantly behind Intel in performance until 1995, when it bought a startup called NexGen. That company's work became the K6 architecture and finally put AMD on equal footing with Intel. The next chip, the K7, would be branded Athlon and build a leadership gap with Intel for the first time.
Designed by ex-DEC engineer Dirk Meyer, who created the 64-bit RISC processor Alpha, K7 incorporated a lot of ideas from Alpha on license from DEC. It was the first x86 CPU to hit 1Ghz, and there would be more firsts. It would become the first dual-core CPU and eventually the first 64-bit CPU. AMD would make a server version, Opteron, and when Microsoft ported Windows Server 2003 to 64-bits, AMD was ready to take advantage of it. It went from no server presence in 2000 to almost 20 percent by 2006 along with almost one-third of the desktop. Intel was left wondering what the hell hit it.
And then it all fell apart. Intel regained its footing and AMD undertook the Barcelona project, which introduced too much change at once. AMD tried to adopt a new microarchitecture, core structure, instructions, interfaces and process technology all at once and couldn't do it. That's why Intel uses the tick/tock strategy: it introduces a new architecture one year, a process shrink the next year. AMD has readily admitted it tried to do too much at once with Barcelona.
AMD also bought ATI Technologies, the Canadian graphics maker and only remaining GPU maker left besides Nvidia. Rumor has said AMD wanted to buy Nvidia, but its CEO Jen-Hsun Huang wanted to run the company, which was a non-starter for AMD. So AMD bought a company across the continent and in another country for the astronomical price of $5.4 billion. AMD would end up writing down about two-thirds of that value over the next five years.
Almost immediately, then-CEO Hector Ruiz set about on a vision he had called Fusion, which would meld the CPU and GPU on one die. Hector's vision was a little too far down the road. It would take many years for Fusion to become a market reality. During that time, AMD lost billions of dollars, market share and support.
But now, seven years, three CEOs and countless millions later, Fusion (renamed Heterogenous Systems Architecture (HSA)) is starting to pay off. AMD's processors are competitive on the low- to mid-range of PCs, but that's not where its focus is. AMD has landed the Sony PlayStation 4, Microsoft Xbox One and Nintendo Wii U. I have never heard of one chip provider supplying all three competitive consoles. Sony and Microsoft chose AMD because an ARM + GPU SoC wasn't up to snuff. In particular, the 64-bit support wasn't there.
One financial analyst predicts the company will earn between $60 and $100 per console. Considering Microsoft and Sony sold 145 million consoles combined over their six to seven year lifespans, if we use an average price of $80 per console and count on Sony and Microsoft matching past success, AMD will earn $11.6 billion just from those two consoles. Then there's the Wii U. If it can match its predecessor, the Wii, that's another 99 million units sold and another $7.9 billion.
That'll pay off that huge debt that's been crushing AMD for years.
So where does AMD go beyond consoles? It's taking a leadership role in the HSA Foundation group, a collaborative efforts to standardize programming an SoC. The goal of the foundation is to hide the hardware complexity from the software and also to make the apps as energy efficient as possible. This is not confined to PCs, as Qualcomm, ARM and Samsung are also members. Intel and Nvidia, notably, are not.
In the process of developing this HSA technology, AMD has made its CPU, GPU and other silicon components so modular it can build SoCs like Lego blocks. Both Sony and Microsoft wanted a custom x86 SoC but they didn't want the same one as the other guy. They needed different graphics, video processing, content security, and memory management because Sony went for raw power while Microsoft wanted to create a living room entertainment console.
Only AMD had the IP and experience to pull it off. It can take a very good CPU and a highly competitive GPU, along with its video, image processing, audio, security, memory controller and I/O and make a variety of SoCs that are competitive and yet different from one another.
AMD is uniquely positioned to dominate the custom SoC market like none of its competitors can at the moment. Intel doesn't have the GPU, but it is closing that gap. Nvidia doesn't have x86. ARM doesn't have 64-bits or a really competitive GPU. AMD is once again positioned to dominate. The real question will be how far down it can push into low-power products to get into the tablet market, which is where the action is these days.
The trick now is to not screw up again like it did with Barcelona.