The chip sports two major features for enhancing performance--, for linking chips, and an for speeding up the data flow between memory and the processor. Benchmarks from AMD suggest that the chip will increasingly outperform Intel's equivalent Xeon chip.
More importantly, the Opteron can run 32-bit Windows-Linux software, which is found on desktops and small servers around the world, and 64-bit software, which is used on high-end servers. The 64-bit software that will run on the chip will be almost identical to the kinds of applications and operating systems corporate customers use now, making it easy and cheap to adopt.
A desktop version of the chip, called Athlon 64, which can be used on anything from laptops to scientific workstations, comes out belatedly in .
Intel chips aren't as ambidextrous. Instead, the company sells Pentiums and Xeons for, respectively, PCs and midsize servers, and it sells Itanium 2 for the 64-bit market. Pentiums and Xeons run the same software, but Itanium 2 requires entirely different code, which can be costly to develop and test. Itanium's software isolation in part has retarded sales, analysts and executives have said, adding to the appeal of AMD's one-chip solution.
"Intel will ultimately have no choice but to respond," said Dirk Meyer, a senior vice president at AMD. "If we didn't exist, I think Intel would be in a position to drive Itanium down everyone's throat. But we do exist, and we are going to offer an alternative, and even a virtual monopolist like Intel is going to respond at some point."
Smaller manufacturers likewill release Opteron servers this week, and analysts believe that big companies like IBM could follow suit. If that happens, AMD will be able to compete against Intel in everything--not just the consumer and small-business segment, which accounts for only around 30 percent of the total market. Microsoft, IBM and Linux companies currently are writing software for the chip.
Then again, this is AMD--a company that seems perpetually stuck in a Sisyphean nightmare.
In the past 16 years, AMD has had eight profitable years and eight in the red. Total net earnings since the end of 1986, including acquisitions and sell-offs, comes to around $350 million, about the same as Intel earns in six weeks.
Bringing Opteron to market hasn't been easy. Both the server and desktop versions of the chip were originally, but they kept getting pushed back, in part because of technical problems. The chip will debut at 1.8GHz, slightly lower than expectations.
Although the company has made its share of mistakes, its biggest problem is Intel, which has more engineers, fabrication facilities, money and connections.
"We only have one competitor, and that one is pretty awesome," AMD CEO Hector Ruiz said in an interview late last year. Intel is "huge, and they have a lot of smart people."
Nonetheless, AMD executives and most analysts assert that the historical pattern is ready to be broken. The company has shed its reputation for haphazard manufacturing and boasts one of the premier semiconductor design teams in the world.
"For the last 10 years they get themselves to the brink of death and then come back. They are great at the last-minute comeback," said Dean McCarron, principal analyst at Mercury Research. "But Opteron is a reliable, corporate-server play. If it gets traction, then everything (AMD makes) is acceptable."
Opteron under the hood
The underlying theme of the Opteron is memory and its problems. Memory, which stores data required by the processors, remains one of the chief bottlenecks in computer performance. Memory runs at a far slower rate than the processor, a mismatch that leaves the processor spinning in idle as it waits for data. The physical distance between memory and the processor also creates a lag.
All three of the touted features of the Opteron work to resolve this problem. Integrating the memory controller effectively shrinks the first part of the pipeline between the memory and the processor, thereby speeding up performance.
Inservers, the central bus is eliminated. Instead, processors and memory are spread out and united by a high-speed ring road, similar to the decentralized live-work areas that have become common in high-population areas.
Additionally, Opteron comes with 16 registers, or data holding bays, instead of eight like existing chips using the so-called X86 architecture, which is the basis for current Intel and AMD chips.
"The X86 architecture is fine, but if it has any one limitation it is in the limited number of registers," AMD's Meyer said. "Going from eight to 16 may not sound like much in comparison to RISC architectures (like IBM's Power 4 chip) with tens of registers, but you get essentially 80 percent of the benefit going from eight to 16."
The reasoning behind Opteron's 64-bit characteristic revolves around memory. Current computers with 32-bit chips can only manage 4GB of memory, while a 64-bit computer can handle vastly more. Databases and other applications perform better with more memory because it allows the processor to access more data rapidly without having to dig it out of the disk drive.
Intel's Itanium provides the same ability, but it's far less convenient and much more expensive, AMD executives say. With Opteron, computer makers and software developers only have to tweak their products modestly to get them to take advantage of the 64-bit characteristics, according to Fred Weber, AMD's chief technical officer. IBM ported its DB2 database over in the course of just a few days with a couple of engineers, he noted.
For corporate customers, this means an easy adoption curve. Opteron also will be far less expensive than Itanium, which starts at $1,338. Adding the 64-bit ability only increased the size of the chip by a small percentage, Weber said. Initial prices for the Opteron range from $283 to $794. That means cheaper hardware.
Large hardware makers are intrigued. Newisys, which is licensing an Opteron-based server design to other manufacturers, says that virtually every major manufacturer is testing the chip.
"The large hardware manufacturers are watching each other to see who bites. If one of them does, the others will be (pressured) to do it as well," said Mike Feibus, principal analyst at TechKnowledge. "I don't see any resistance at the (information technology) level. It is all with the computer makers."
Intel, of course, isn't taking the challenge lying down. The company will upgrade its Xeon processor line in the near future and has been working to popularize its Itanium 2 chip. Richard Wirt, an Intel fellow and one of the company's senior chip designers, said he meets with researchers and developers and Oracle on a weekly basis to optimize performance. Overall, Intel has spent millions on rounding out the corners on Itanium 2.
And although performance of the first Itaniums was disappointing, recent benchmarks have put Itanium 2 toward the top of the heap. A new version, code-named Madison, will further boost output.
Sales have picked up as well. Oil giant BP, Swiss laboratory CERN and a number of research institutions have adopted Itanium. Dell Computer, an early critic of the chip, has said it will sell.
Additionally, AMD has a sordid history to live down. Although it's shown it can create award-winning chips, making them in volume profitably has proved elusive.
Compaq Computer shocked the PC world in 1995 by announcing its intent to adopt AMD's, a decision it reversed after performance problems emerged. With the K6 and K6-II, AMD gained market share and simultaneously lost billions.
In mid-1999, thewon rave reviews and contracts with Hewlett-Packard, Sony, IBM and Gateway. Two-and-a-half years later, the company has about 13.8 percent of the processor market, or around the same amount it did before the Athlon arrived. Although the self-imposed problems have faded, Mercury Research's McCarron and others point out that big computer companies are conservative to a fault and rarely like to take risks.
Even if AMD can manufacture flawlessly however, the company will face another obstacle: indifference. While 64-bit performance sounds great on paper, actual demand in the broad market could be far off. In the server market, more than 80 percent of the servers sold are 32-bit machines. Most Opteron servers sold, therefore, will be basic 32-bit machines.
In desktops, 64 bits may not be in the picture for years. Memory can, at most, account for 8 percent to 10 percent of the cost of building a PC, according to memory and PC executives. Although memory prices are dropping fast, 4GB of memory is more expensive now than many desktops. High-end PCs with 4GB of memory will still be rare in, according to Dataquest analyst Andrew Norwood. These computers won't even be mainstream by 2007, according to Sherry Garber, an analyst at Semico.
By that time, AMD will be selling new chips, and the current Athlon 64 and Opteron designs will be a distant memory. Intel executives have said they probably won't bring 64 bits to the desktop until near the end of the decade. Besides, the company has technology that can let 32-bit computers exceed the 4GB limit, although not to the same degree as 64-bit chips.
In the end, what AMD will likely need is a computer maker to carry the 64-bit flag.
"What is absolutely critical is to get those top-tier manufacturers," said Gordon Haff, a senior analyst at Illuminata.