AMD releasing details on 64-bit Sledgehammer
Some of the mystery surrounding Advanced Micro Devices' processor will be lifted when the company posts a technical manual geared at recruiting software developers to the project.
The document, called the Architecture Programmers' Overview, explains the technical principles underlying Sledgehammer, the successor to the Athlon processor. Sledgehammer is slated to appear in samples by late 2001. Commercial shipments will likely start in 2002, according to AMD.
"The higher-level architecture is done," said Bob Mitton, division marketing manager at AMD. "We want to start to roll it out and get everyone to look at it."
Sledgehammer is one of the company's most ambitious projects to date. The chip will run software designed for 32-bit processors, such as Intel's Pentium III and AMD's Athlon, as well as software for 64-bit chips, a class of processors that includes Sun Microsystems' UltraSparc, Compaq Computer's Alpha and Intel's upcoming Itanium.
The 32- and 64-bit designation refers to the amount of data that a processor can handle in a single instruction: The larger the number, the more data that can be processed at one time. With 64-bit chips, server makers also can incorporate much more memory in their machines.
Itanium and Sledgehammer differ in the type of 64-bit programs they will run. For Itanium and its successors, Intel and Hewlett-Packard developed an entirely new instruction set, or processing protocols, for the chip. Software is being developed to recognize Itanium's instruction set.
By contrast, Sledgehammer will use the same basic instruction set as the Athlon but will contain features that allow the chip to run 64-bit programs. As a result, software makers only need to retrofit their programs, not write completely new ones.
"This is the conservative step approach. It is not a major disruption," Mitton said. "If it's not broke, why fix it?"
Similarly, Sunnyvale, Calif.-based AMD will only need to concentrate on one chip family. The underlying core of Sledgehammer chips in desktops, notebooks and servers will essentially be identical, Mitton said. It will be up to computer manufacturers to decide whether to have a given processor run as a 32-bit or 64-bit chip when they assemble their machines. The core of Intel's 64- and 32-bit chips will be different for years.
The question, of course, is whether the chip will prove to be effective in both types of machines. Making a chip ambidextrous almost always creates a performance hit.
With Itanium and its successors, Intel is aiming to provide optimal performance with 64-bit chips. These processors will run standard 32-bit applications, but with no performance enhancements.
The big issue for Sledgehammer is how well it will run 64-bit applications. The chip will likely perform fine in desktops, but adoption by server companies may be slow if Sledgehammer can't approximate Itanium's performance.
"With compatibility, the positive is that you are bringing everyone along. The negative is that you are carrying baggage," said Mike Feibus, principal analyst at Mercury Research. AMD also faces the challenge of keeping costs down.
In some ways, though, AMD can't lose, said Insight 64 analyst Nathan Brookwood. If it works, AMD will gain entry into the market for high-end servers without having to spend huge amounts in research and development. If it doesn't work, the company will have a fast desktop processor.
AMD will also potentially enjoy an advantage in convenience. Migrating to a new instruction set is painful, difficult and expensive for everybody, including the chipmaker and software developer.
"As long as you maintain binary compatibility, it becomes a no-brainer because there are no software compatibility issues," Brookwood said.
AMD, in fact, won't be the first company with an ambidextrous chip. In 1985, Intel was working on a 32-bit chip called the 432 to replace the 16-bit 286. The 432 contained an entirely new instruction set.
"It was going to be a really nifty machine. All the serious architects were working on it," Brookwood said.
Intel, however, killed the project and instead released the 386, a 32-bit version of the 286 that also read 16-bit code. One of the chief engineers on the project was Pat Gelsinger, who has since become Intel's chief technical officer.
"The second string became heroes," Brookwood said.