AMD ready to drop the Hammer with Opteron launch

Advanced Micro Devices Inc. (AMD) has made itself into a consistent thorn in Intel Corp.’s side with its Athlon XP desktop processors over the last few years, but hasn’t made much headway against Intel’s popular Xeon processors and notorious Itanium 2 processors for servers. On Tuesday, the company will unveil its Opteron server chip, which puts a new twist on a familiar technology.

The Opteron’s launch event in New York comes after almost a year’s worth of hype and promise from AMD. Based on the company’s Hammer architecture, the processor uses 64-bit extensions to the x86 instruction set to allow users to run both 32-bit and 64-bit applications on the same chip. Chips for one-way, two-way, and four-way servers will be available on April 22, said Marty Seyer, vice-president of marketing for AMD’s server business, during an interview at Linuxworld in January.

A 64-bit processor uses wider general purpose registers than a 32-bit processor. A general purpose register is a place on the processor where data is temporarily stored as it is processed, a holding spot for data before it gets moved to another space within the processor.

The wider general purpose register can allow up to twice as much data to be processed by a 64-bit chip in the same clock cycle as a 32-bit processor, said Kevin Krewell, senior editor at the Microprocessor Report in San Jose. It can accommodate a 64-bit wide instruction, whereas a 32-bit processor would need two clock cycles to process that same instruction, he said.

Database managers also like 64-bit processors because they can address more memory than 32-bit processors can handle. Current 32-bit systems are hamstrung by a 4GB limit in the amount of RAM a system can address at any one time. This isn’t a problem for most desktop users, who rarely need that ability to run current applications. However, servers with 64-bit chips can hold enormous amounts of data in larger memory banks, which increases the speed at which that data can be processed.

Databases will be the primary application for the Opteron, at least at first. At Comdex last November, IBM Corp. and AMD showed a demonstration of IBM’s DB2 database software running on the Opteron, and IBM announced beta availability of that database for the Opteron at Linuxworld.

Microsoft Corp. is expected to release a version of SQL Server for the Opteron, Krewell said, coming off that company’s announcement that it will release a version of Windows Server 2003 for the chip. Microsoft’s recent decision to make its current 64-bit version of SQL Server the same price as the 32-bit version makes it an attractive proposition for IT managers that already use the 32-bit version and are considering the Opteron, Krewell said.

“They can get more benefit out of SQL Server without it costing anything else,” he said.

Oracle Corp. has said nothing yet about the Opteron, and that could present problems to widespread adoption of the chip, Krewell said. “Oracle is a very influential database company, they are the status quo. But the (database) companies that want to grow market share more will jump on the AMD bandwagon,” he said.

Companies such as Sun Microsystems Inc. and IBM Corp. have been shipping 64-bit reduced instruction set computing (RISC) processors for many years, but Intel and AMD each are taking different approaches to 64-bit computing than their Unix counterparts.

The x86 instruction set has provided the marching orders for most PCs, and server processors such as Xeon, for about 20 years. Intel Corp. developed the architecture in the early 1980s, and both Intel and AMD continue to use it today in their processors. Because of its history, and close association with Microsoft Corp.’s Windows operating system, thousands of companies use applications written for the x86 platform.

But with the release of the 64-bit Itanium processor in 2001, Intel bet that its new explicitly parallel instruction computing (EPIC) architecture would offer enough performance benefits to justify the effort companies would have to make to port their applications from 32-bit server processors or 64-bit RISC processors over to the new instruction set. Opteron and Itanium are not expected to compete head-to-head, at least initially, but represent different philosophies on how to move enterprise customers to 64-bit capability.

When it was designing the Itanium processor family, Intel looked at several existing architectures, and decided none of them would be adequate to support the performance goals the company had for Itanium, said Barbara Grimes, an Intel spokesperson. “We wanted to deliver more than just memory capacity, we wanted to deliver reliability, availability, and serviceability, which you don’t get from just moving to 64-bits,” she said.

“Intel was looking out 15 to 20 years, and they did not feel that doing an AMD-style extension was the way to get that kind of longevity. From a purely technical perspective, there is a lot of baggage in the x86 architecture. If you were designing a chip today, there’s no way you’d put many of those instructions in a chip,” said Gordon Haff, an analyst with Illuminata Inc. in Nashua, N.H.

However, IT managers outside of the high-performance computing market have been reluctant to move their existing applications to Itanium, Haff said.

The problem with EPIC is not porting a single application, it’s porting all of a company’s applications, Haff said. “All your software needs to have unique Itanium versions. It’s not that there is great difficulty supporting that software, but you need to create those new versions, test them, and certify them for your systems on a platform that hasn’t seen much traction outside of high-performance computing,” he said.

This leads to an old-fashioned chicken-and-egg dilemma, Haff said. “People won’t buy (Itanium systems) because there aren’t a lot of applications, and there aren’t a lot of applications because there hasn’t been much interest (in those systems),” he said.

AMD hopes that its approach to 64-bit architecture provides IT managers with a way to keep their current applications running while developing or testing 64-bit applications.

“With Opteron, not all applications have to be brought over to 64-bit, some can stay right where they are,” Seyer said.

Intel enjoys a solid position in the server market with its 32-bit Xeon processors, but AMD is trying to convince users of low-end servers to move up to a 64-bit chip that can support their critical 32-bit applications while they recompile their code to take advantage of the 64-bit register widths, Seyer said.

Rumours that Intel is working on its own x86-64 chip have persisted ever since AMD announced its plans, but Intel has steadfastly denied the existence of “Yamhill,” the purported code name for the chip. Many analysts and observers feel that if AMD’s approach is successful, Intel will have to come out with its own x86-64 chip in order to maintain its server market share.

AMD wants to eventually compete at the high end of the market, against Itanium and expensive RISC systems from Sun, IBM and Hewlett-Packard Co., Seyer said. RISC chips decrease the number of instructions the processor needs to understand, but increase the number of times those instructions are carried out. Chips built on the x86 architecture are known as complex instruction set computing (CISC) chips.

However, AMD’s niche will come in the Xeon space, and the company won’t gain much traction in the powerful back-room servers for a while, Haff said.

“This is a company with virtually no track record in the enterprise. We’ll see them on the desktop and dual-processor servers, not really competing with RISC,” he said.

The Opteron chip suffers from a lack of floating-point unit performance, a strong suit of Itanium and RISC chips, Haff said. A floating point unit is a special set of instructions that processes complex numerical data, and is important to life sciences and other mathematics-intensive applications.

None of the major so-called tier-one server vendors, such as Hewlett-Packard Co., Dell Computer Corp., or IBM Corp., have said they will use Opteron in their products, but AMD hopes to build market share through white box vendors.

White box vendors are generally small, local firms that build and maintain servers without a well-known brand name. Individually, they can’t compete against the tier-one players, but collectively they shipped the second largest number of servers in 2002, according to Dataquest Inc.

AMD will remain committed to its Cline of servers through at least the end of the year, but the pricing might change as AMD decides how much to charge for Opteron, said Seyer.

AMD is on the Web at