Opinion: Is AMD Doomed?

By Loyd Case  |  Posted 11-19-2007

Opinion: Is AMD Doomed?

I was as startled as anyone to see last Friday's news that Abu Dhabi's Mubadala Development Company handed AMD $622M in exchange for 8.1% of the company. That puts AMD's overall valuation at roughly $7.7B, slightly above what the current stock price would indicate.

That $622M infusion gives AMD some vital financial breathing room, given that it's lost more than $1.6 billion so far this year. Add $5.3B in debt, and only $1.5B in cash on hand (now, presumably, $2.16B), the situation is not pretty.

The real heart of the matter, though, is not AMD's financial situation. The company's financial situation is directly the result of its execution in the market.

Let's step back a few years, to AMD's halcyon days. Back then, CTO Fred Weber was on a roll, after having architected AMD's rise to performance prominence over archrival Intel with AMD's Hammer processor family. K8 became known as the Athlon 64 and later, the Athlon 64 X2 as AMD shipped dual core versions. The native dual core X2 CPUs were the darling of PC performance enthusiasts everywhere, running rings around equivalent Intel processors in terms of performance while at the same time using less power and putting out lower heat.

Intel's only response at the time, in terms of product, was to slap two very hot CPU dies into one package and call it a dual core CPU. On the one hand, they actually beat AMD to market with their dual core Smithfield product. But it was a very hot solution, and was an underperforming relative to what AMD was shipping.

AMD was watching this, and decided to roll the dice: its next generation CPU would be a "true quad core" processor, with all four cores on one semiconductor die. After all, it crushed Intel in terms of performance when Intel tried to put two dies on one package. Even if Barcelona, as the new architecture was dubbed, was late, AMD would still be able to put the hurt on Intel.

What AMD apparently wasn't paying much attention to was this little product line called "Centrino."

In truth, Centrino was a lot of marketing hot air—but very profitable marketing hot air. The whole platform approach obscured the fact that underlying Centrino was a pretty killer mobile CPU, the Pentium M. It was performance efficient, offering much higher instructions per clock cycle than Intel's desktop CPU, and it was energy efficient, consuming relatively low power. At its original inception, the Pentium M was something of a stepchild inside of Intel, having been developed outside of Intel's mainstream architectural design process.

But the success of Centrino and the Pentium M made the powers that be sit up and take notice. Intel rolled some dice of its own, and decided to turn the aircraft carrier around. It cancelled existing projects, like the Netburst-based Tejas CPU, and began architecting a new CPU based on the design principles of the Pentium M. All this was a huge vindication for Intel executive Mooly Eden and Intel's Israeli design team.

But enough about Intel. The key point here is that Intel realized what it had in the Pentium M—though it took some time—while AMD didn't seem to quite realize the threat. Intel was big, cumbersome and constantly talking about different architectures. For all AMD knew, Intel's Conroe project would just be another underperforming Intel CPU.

To be fair, AMD's resources are more limited than Intel's, given the much smaller size of the company. Until recently, AMD has proven to be as efficient at husbanding its resources as its Athlon 64 X2 CPUs had been. So AMD focused on designing Barcelona, bringing up new fabs and churning out faster iterations of its existing CPU line—as you'd expect a CPU company to do.

Then like a high roller in Las Vegas, AMD returned to the craps table and made a huge bet: it bought Canadian graphics hardware company ATI Technologies. Continued…


Quite a few analysts were baffled by this. Some synergies are apparent: AMD really needed its own native core logic, instead of completely relying on third parties like Nvidia and Via. But buying ATI earned the enmity of one of its key partners—Nvidia. It also took on substantial debt to acquire ATI, although it did seem like ATI was riding high at the time. Both ATI and Nvidia had been battling back and forth in the GPU market, with no clear leader emerging.

AMD believed it could integrate key aspects of GPU technologies into future CPUs. Once again, AMD was proving to be more forward looking than the competition. But it was also diluting its focus, having to keep more balls in the air than ever before.

Then in early 2006, rumors were starting to float around the ether about the performance of Intel's Conroe. The initial rumors were purely unbelievable, given Intel's history with Netburst. Later, those early numbers seemed like understatements, after Intel launched the Core 2 product line in mid-2006. Core 2 not only outperformed AMD's best CPUs, but did so at lower clock frequencies and lower power draw. Intel's full weight had come down upon the Hammer.

AMD did what it could, but the company's foray into 65nm process technology didn't buy as much headroom with the Athlon 64 as it hoped. And Barcelona kept receding into the horizon. Worse, the graphics division's new DirectX 10 GPU was also late, running into some very pesky and hard to debug teething problems, even as Nvidia shipped its DirectX 10 8 series GPUs. So AMD has been forced to play the pricing game—and started down the road of huge financial losses even as it sold increasing numbers of processors. Sell one at a loss, make it up in volume, seems to be the watchword.

AMD finally shipped server-based Barcelona CPUs in Q3 of 2007 and is now readying some desktop Phenom processors. But as you can see from ExtremeTech's Phenom performance preview performance is lacking, once again forcing AMD to play the pricing game. A new flagship CPU should be the margin leader, but Phenoms build on 65nm offer too little performance and run too hot. The company that once pounded the CPU efficiency drum has been relegated to discussing the low power usage of its chipset.

And while AMD has been talking up its 45nm process technology, it's increasingly looking like it will finally be able to build 45nm Phenoms just as Intel is shipping its brand new Nehalem architecture.

So far, it seems like a lot of gloom and doom is all we can hope for regarding AMD. But there are some bright spots. The new 3800 series GPUs still lag behind Nvidia in terms of raw performance, but for many users, will have enough performance. The die size is exceptionally small for such a high transistor count chip, giving it a potential cost edge over Nvidia. At the same time, Nvidia is having great difficulty building enough 8800 GT's to meet demand.

The new 790FX chipset looks very promising, offering a more flexible approach to multi-GPU support than Nvidia and using less power than its completion. The new Firestream 9170 ups the ante in the GP-GPU arena, taking on Nvidia's Tesla product line and adding the new wrinkle of double-precision floating point support. (But the 9170 isn't shipping until the end of Q1, giving the competition plenty of time to respond.)

So can a $622 million dollar cash infusion plus reasonably profitable GPUs give AMD enough breathing room to prove its gamble on true quad core was a good one? The jury is still out, but the portents are not very encouraging. And that would be too bad, because the tech world really needs a scrappy, nimble CPU company like AMD, just like it needs a scrappy, cool computer company like Apple.

Can AMD survive long enough to pull itself up by its bootstraps? Drop by the forums and tell us what you think.