AMD’s 64 Bit Gamble

Pages: 1 2 3 4 5 6

The 64 Bit Question: Why?

Ironically 64-bit processors are used in both high-end technical and commercial computing applications, and in cost sensitive embedded control applications that rely on the CPU to do heavy data movement and manipulation in lieu of specialized hardware (although the processors for these two markets differ tremendously in cost, performance, and power consumption). The majority of desktop computing is performed with 32-bit processors, namely x86 compatibles (and to a much smaller extent, PowerPC processors). Advanced Micro Devices sells the vast majority of its MPU products into the 32-bit desktop market and very little into the two areas where 64-bit processors are dominant.

The obvious question is, why does AMD feel the need to extend the x86 architecture to 64 bits when there is little need for it in the markets AMD serves now and over the next few years? Part of it is simply that neither Moore’s law, that predicts the rate of improvement to IC technology, and Gate’s law of ever bloating application and system software, show any signs of slowing. Today it is not uncommon to see personal computers outfitted with 256 Mbytes or more of memory. Although the 4 Gbyte memory addressing limit of 32-bit processors won’t be within the budget of most people for quite a few more years, it will soon start to become a concern for the bleeding edge, high end PC users that AMD wants to continue to sell x86 compatible processors to.

The second reason is Intel’s slow but inexorable march towards replacing x86 with 64-bit IA-64 processors. The Merced/Itanium will likely be too clumsy, hot, and underpowered (it reminds me of the 0.8 um Pentium) to make much of an impact in the face of the Willamette x86 processor, and McKinley will probably be too expensive for the desktop. But it is rumored that Intel has more than half a dozen separate IA-64 processors currently under development. According to Intel, IA-64 processors won’t ‘officially’ target the desktop market until the price/performance oriented ‘Deerfield’ processor ships sometime in 2002. Anyone with a bit of history under their belt remembers how each new Intel processor generation from the 80386 onwards was accompanied by industry analysts pronouncements how it was a ‘server chip’, it was ‘overkill for the desktop PCs’, users would ‘never need the power of the new chip’ etc. The history lesson from that era is that Intel can drive a new processor core to the mainstream remarkably fast, especially if there is effective competition against the previous generation.

Given the fierce competition Intel faces in the x86 marketplace it would be very tempting to try to transition the high end of their customer base to IA-64 as quickly as possible. But low to mid-range IA-64 devices will not likely offer much performance advantages over future shrunk versions of the Willamette core and their AMD counterparts, especially on integer applications. What does Intel do in such situations? It tries to generate demand for a new proprietary technology through an intense and focused marketing campaign. The two areas where IA-64 will stand out from x86 are in its ’64-bitness’ and its floating point performance. As a result, those would be the likely axes of attack for an ad blitz to convince PC buyers that they had to upgrade to IA-64. Whether or not these attributes are actually meaningful for the end user is immaterial (remember the Pentium III sales campaign that tried to sell the laughable notion that SSE would greatly improve the user’s Internet experience?).

The x86-64 and TFP enhancements to the AMD K8 seem to be a clear attempt to limit the marketing leverage that Intel can exert when it finally attempts the transition from x86 to IA-64. AMD has no choice but to try to keep x86 viable as long as possible. The technical and legal challenges that stand in the way of cloning IA-64 are very large. Even if it was feasible to clone IA-64 it would be a distasteful retreat to the past for a company that is justifiably proud of its independent engineering achievements in bringing the K7 Athlon to market and beating Intel at its own performance and clock speed game.

Pages: « Prev   1 2 3 4 5 6   Next »

Be the first to discuss this article!