AMD's best is yet to come
- — 28 August, 2006 15:33
A reader recently shot me a note saying that after studying some of my work related to Advanced Micro Devices, he has spotted a pattern: I always side with the underdog. This reader crystallized the prevalent viewpoint, one expressed by most of my colleagues, that AMD achieved its market success by dumb luck, with the "dumb" supplied by Intel. Intel discovered too late that NetBurst was, well, everything I've always said it is, and hubris born of market domination left Intel without a Plan B.
Now that Plan B (Core Microarchitecture) Xeon CPUs and chip sets are here, I'm hearing that the x86 market has returned to its natural balance. Intel is once again the exclusive synonym for x86, and AMD is again relegated to the fringe. It is the low-cost or counterculture alternative to the one true architecture. This fallacious mob certainty was reached before AMD brought the next-generation Opteron to market, making clear how easy it is for Intel to lead the market around by its nose.
The bizarre twist is that everyone knew precisely what AMD was doing next in server CPUs. Revision F Opteron, now called next-generation Opteron, has been a known quantity to my readers for a long time. It virtualizes and scales beautifully from entry to enterprise. Next-gen Opteron uses the same RAM as that employed by Core Microarchitecture, but Opteron's memory controllers are on-chip; Intel's are external. Intel still uses a front-side bus architecture in which memory, inter-processor and peripheral data compete. Core Microarchitecture now uses two such buses, but broken plus broken does not add up to breakthrough.
Study Intel's rhetoric closely: Core Microarchitecture's big numbers, like the "40 percent faster," represent Plan B's lead over Netburst Xeon in cycle-per-cycle performance and power efficiency. Intel has indeed left its prior best in the dust, but Intel's presentation of this fact is almost indiscernibly blended with Intel's far less impressive claimed lead over Opteron. It's working: There are a great many smart, respectable IT folk who think that Core Microarchitecture has a massive lead over Opteron.
Here's the truth: The direct performance-per-watt numbers that Intel has published to bolster its claims of lower power usage pit a high-power Revision E Opteron against a low-power Core Microarchitecture Xeon. AMD ships 35 and 55 watt Opteron CPUs and always has. AMD's PowerNow! run-time power management has been standard in Opteron for a long time; it is not Intel's invention.
Intel shot its entire wad on Core Microarchitecture. From here, the only place Intel can go is bigger cache, more cores and faster clocks. That sounds like a grand triple play, but it isn't. Mark my words: Core Microarchitecture will not scale. When quad-core Opteron lands, it will be ready to power 32 core-capable servers that deliver close to linear performance gains over Opteron systems with fewer cores. Plan B Xeon is entry-level server technology and even in that category, AMD has the better story.
AMD's got years worth of ammunition already locked and loaded. It hasn't even played the 65-nanometer manufacturing process card, one that Intel had to play just to get Core Microarchitecture out the door. I'm certain that AMD is truly ready for 65-nanometer and other mind-blowing things, and I'm just as certain that Intel's claims of Core Microarchitecture's technological lead over Opteron will prove baseless. AMD's road map is guided by IT's needs and the capabilities of enterprise applications, rather than Intel's provocations.
