High-End Processors: Are They Worth All the Fuss?


It's no surprise that chip sales are down considering the havoc the economy has already wreaked on the server and storage industries. The real question is why AMD and Intel continue to devote so much of their dwindling resources into developing high-end processors.

iSuppli just came out with its first-quarter microprocessor market figures, and as we've all grown accustomed to by now, the numbers are not good. Overall revenues are down more than 20 percent compared with a year ago, with only low-end, notebook processors like the Atom showing any signs of life at all. The overall market, which includes everything from x86 to RISC processors, dropped from $8.6 billion in the first quarter last year to $6.9 billion in the first quarter of 2009.

Nonetheless, it was a sweet victory for AMD, at least, in that the company saw its market share inch up a bit vs. Intel. AMD's revenues contributed 12.8 percent of the overall market, a 2.3 percent gain, while Intel lost 2.5 percent to come in at 79.1 percent. While Intel continues to dominate, it is the first loss of market share for the company in four quarters. Keep in mind, though, that it still represents a net loss in sales for both companies.

While all sectors of the microprocessor industry are struggling, the high-end devices are placing the largest drag on the market. With capital budgets at top enterprises essentially frozen due to the recession, and with the success of hardware-extending technologies like virtualization, it's no surprise that top-end devices like the multicore Xeon and Opteron lines are seeing lean times.

That's why it's puzzling to see both companies waging a pitched battle in the high-end space. AMD recently introduced its six-core Opteron, the Istanbul, said to outperform the current top-of-the-line Shanghai chips by some 30 percent while keeping the same power envelope. The chips are designed for up to 8-socket servers and offer 6 MB of L3 cache and 512 KB of L2 cache per core. And by next year, the company is eyeing a 12-core device called the Magney-Cours.

These chips are intended to counter Intel's efforts in the race to load up on cores. The company is gearing up for the eight-core Nehalem EX, a 45nm device that supports up to 16 threads and 24 Mb of cache and is loaded with more than 2 billion transistors.

Clearly, these devices are aimed at weaning high-performance computing (HPC) shops off of their RISC/mainframe platforms and onto x86 clusters, but that could prove to be a risky gamble. Once the economy pulls out of recession, there will likely be an uptick in activity, but it could be quite a while before a real market re-emerges. IBM is still maintaining a comfortable mainframe business, and the technology, old as it is, shows a lot of promise in new cloud-based infrastructures. And besides, wasn't usurping the mainframe the rationale behind the Itanium?

To be fair, both companies are aggressively pursuing the laptop and mobile market: Intel with the Atom line and AMD with a new family of Neo processors aimed at the ultrathin market. But even though the margins on these lower-end devices can't possibly be as good, there's something to be said about pursuing opportunity wherever it leads.

And from the view in June 2009, it seems that expensive core enterprise systems have had their day. Low-cost endpoint hardware is where the action will be for some time to come.