In-Depth

AMD Dual-Core Lead Leaves Intel Far Behind

AMD has forced Intel to abandon its decades-long marketing strategy, leaving the chip-making leader struggling to find its sea legs

• By Stephen Swoyer
• 05/17/2005

Advanced Micro Devices (AMD) continues to make progress at Intel Corp.’s expense, and the situation could escalate over the next several months. Intel won’t ship its first dual-core processors for servers until early 2006; AMD is shipping dual-core versions of its 64-bit Opteron chip now.

According to market watcher Mercury Research, AMD controlled 16.9 percent of the worldwide PC (x86) market during the first quarter of 2005. Although AMD’s showing is still dwarfed by that of Intel (with an 81.7 share), it’s still much improved (by 1.9 percentage points) over its year-ago quarter. Intel, by contrast, saw its own share of the worldwide x86 market slip by about 1.8 percent. The upshot, Mercury analysts say, is that AMD’s gains came almost entirely at Intel’s expense.

Intel’s Blues

It’s been a rough year for Intel, analysts agree. In the consumer space, the company was forced to back away from an aggressive roadmap that would have taken its Pentium 4 processors to 4 GHz and beyond. The culprit? Thermal dissipation issues attributed to the Pentium 4’s already-high clock speed.

As a result, some industry watchers suggest that Intel’s existing P4 architecture is effectively maxed out. Relief will come in the form of dual-core processors, to be sure – but (for servers, at least) not until next year.

Multi-core processors are old hat to many in the server market. Vendors such as IBM Corp. (Power), HP (PA-RISC), and Sun Microsystems Inc. (UltraSparc), among others, all offer multi-core support, and Big Blue actually supports as many as four cores on a single Power5 chip.

Multi-core in a market that depends on sales volume is another matter entirely. This year, AMD became the first volume player to enter the multi-core club, with its Opteron chip for servers. Intel now expects to deliver a dual-core design for servers by late this year or early next year, but its more sophisticated dual-core chips aren’t expected until late 2006 at the earliest.

“Intel clearly has put all of its bets on dual core, and in the server space, it hasn’t gotten there yet,” says Nathan Brookwood, a principal analyst with microprocessor consultancy Insight64. “So it’ll be sometime early in ’06 before end users can buy Intel-based servers with dual-core processors, and AMD is there now. For some [customers] that could make a big difference.”

This coup, along with steadfast execution on the desktop, have been a boon to AMD, analysts say, helping to create much-needed visibility, even if they haven’t resulted in explosive sales. At the very least, they’ve also caused marketing consternation for Intel.

“[Intel] certainly seemed to think that they could get away with just cranking up the gigahertz for a good bit longer rather than going to multi-core, and all that kind of came together as almost sort of a perfect storm of sorts last year for Intel,” agrees Gordon Haff, a senior analyst with consultancy Illuminata. “They suddenly couldn’t keep cranking up the clock rate, and other vendors were starting to do multi-core. At the same time, AMD had first of all a very nicely performing server processor, and secondly, one that had 64-bit extensions, a path that Intel had decided not to follow.”

The emphasis there, of course, is on the past tense. Early last year, in a highly public mea culpa, Intel came clean with an announcement that most folks viewed as inevitable: It had developed and would deliver 64-bit extensions for its 32-bit x86 server- and consumer-oriented chips. It was hard to see this as anything but a blow to Intel’s flagship 64-bit Itanium processor, which throughout 2003 was touted as Intel’s only on-ramp to 64-bit-land.

“What has happened over the last year is that primarily HP and Intel, they’re basically the drivers of the Itanium bus. [They] have pretty clearly laid out where Itanium plays and where it doesn’t really play,” says Haff. “At this point, they seem to have pretty much ceded the low-end 64-bit space, where maybe they once thought Itanium could ship in volume, to the x86 64-bit systems.”

Storming the Castle

No factor contributed more to Intel’s decision to extend 64-bit support to its x86 line than AMD’s Opteron chip, which—thanks to OEM deals with HP, Sun, IBM (to a very limited degree), and others—competes with the chipmaker’s Pentium 4 and Xeon chips in the low- and midrange server spaces.

AMD, for example, claims that 55 percent of Global 100 companies have opted for Opteron. That’s up from 40 percent last year, and is sharply improved from AMD’s infinitesimal server market share before that.

In the past, AMD’s microprocessor offerings simply haven’t made inroads into the backroom, in part because they lacked robust, scalable, and reliable chipset logic—at least, compared to what Intel itself offered. In designing Opteron, however, AMD obviously took this into account. Their chip features an integrated memory controller and support for high-speed, point-to-point I/O (via AMD’s HyperTransport interconnect) between the processor and system bus.

As a result, analysts say, AMD’s historical difficulties in the server space have largely been mitigated: there’s very little chipset logic for motherboard makers or OEMs to develop on their own, and I/O traffic—the lifeblood of SMP servers—zips along on a fat 12.8 GB/s pipe.

“This helps help AMD for a number of reasons,” says Haff. “First of all, it gives good performance, another area that Intel is still probably okay with, but, again, [Intel has] not moved to serial connections around the processor pretty quickly, and that at least starts to become a performance limitation.”

Secondly, says Haff, as a server processor, Opteron simply delivers the goods—at least relative to its x86 competition. “Certainly by using those high performance [HyperTransport] links, by embedding the memory controller in the processor itself, AMD has a very nicely performing chip that can be built into very nicely performing systems, without depending on any third parties to make a good supporting chipset for them,” he concludes.

Fright Night

Just how much of a scare has AMD thrown into Intel? The Sunnyvale, Calif.-based chipmaker has forced its rival to fundamentally alter its Gigahertz-centric marketing strategy. To a large extent, the Pentium 4 (and its server-oriented cousin, the Xeon) were propounded on the basis of such an approach: Both chips tap a longer-than-usual pipeline, which, clock-for-clock, doesn’t perform as well as AMD’s Athlon, Athlon 64, or Opteron chips—nor, for that matter, the Pentium III chips they were designed to replace.

Intel settled for this trade-off because it expected to ride the Pentium 4 to insanely high clock speeds—four times (or more) as fast as the Pentium III, which started to max out at about 1.4 GHz. In retrospect, the Pentium 4 and its Xeon-based cousin may have been something of a miscalculation, says Haff. If nothing else, he suggests, both chips mark the end of an era for Intel, which has since abandoned its clock speed-centric marketing in favor of an approach, based on discrete model numbers, similar to that of AMD.

“The Pentium 4 in retrospect was probably a sub-optimal approach—that’s a fair statement,” he says. “It was certainly, at the very least, the last chip to take that ultimate pipelining, ultimate gigahertz-type of approach.”

So will Intel’s setback translate into enormous market-share gains for AMD? It depends on how you look at it, says Insight64’s Brookwood. In the most recent year-over-year example, Brookwood points out, AMD increased its overall share to 16.9 percent – a 13.3 percent improvement over its standing (15 percent) in the previous year.

Intel saw its share slip by 1.8 percentage points – or slightly more than 2.1 percent over its standing in the previous year.

At this stage of the game, Brookwood says, it’s not exactly a zero-sum proposition for either competitor. “Intel has far less pain in terms of the revenue that it’s not getting compared with the benefit that AMD has with the revenue growth its seeing,” he says. “I believe there are going to be many Intel customers who, if they have been happy with Intel and are nervous about AMD (and it looks like it’s a six months issue [until dual-core]), they’ll probably continue with Intel. In that regard, I don’t think Intel has a tremendous amount of vulnerability.”

Long-term, Brookwood predicts, AMD will continue to make inroads at Intel’s expense. Of course, just how far AMD will go is anyone’s guess. At this point, the world’s number-two PC and server vendor—Dell Computer Corp.—is still an Intel-only play. “I don’t even think AMD sees its market share growing beyond 50 percent, but I do believe they are likely to continue to make inroads against Intel in terms of market share,” he predicts.