AMD Wields Hammer In 64-Bit Computing

It's an old game. Intel ships a new x86 CPU, AMD clones it and fights Intel for market share.

But with the advent of Intel's new 64-bit Itanium processor, based on Explicitly Parallel Instruction Computing (EPIC) technology jointly developed by Hewlett-Packard and Intel, Advanced Micro Devices Inc. has had to create its own niche for 64-bit computing.

AMD's answer to Itanium and to Intel's future IA-64 chips is the x86-64 architecture. The key difference between AMD's x86-64 architecture and Intel's IA-64 architecture is that while x86-64 is a direct extension of the x86 line, the IA-64 is an entirely new architecture.

By designing a 64-bit chip that's an extension of the 32-bit x86 architecture, AMD hopes to bring to market a chip that's less expensive and offers better backward compatibility with the vast number of existing systems that customers have in place.

For 64-bit customers, AMD's chip could mean less-expensive systems and a clear migration path that preserves investments in existing 32-bit operating systems, applications, and device drivers.

Hammer is the first CPU in AMD's x86-64 line. AMD expects samples of the chip to be available in the first half of next year and systems to ship with the CPU in the second half of 2002.

The Hammer line includes the ClawHammer for single-processor desktop systems and the SledgeHammer for servers and high-end workstations.

Why didn't Intel just extend the x86 instruction set to 64 bits as AMD has done? The primary reason is performance. The IA-64's EPIC technology offers a higher degree of parallelism and future scalability than was possible by extending the existing x86 architecture to 64 bits.

The price Intel paid to implement EPIC was the degree and quality of backward compatibility with its existing x86 chip line. While Itanium has binary compatibility in hardware for the IA-32 instruction set, it runs 32-bit binaries significantly slower than they would run on 32-bit processors.

In contrast, Hammer will run all 32-bit x86 operating systems and applications faster, according to AMD. "The absolute primary advantage is compatibility," says Fred Weber, AMD's engineering VP. "If you buy a computer with the Hammer chip, you can run all the 32-bit operating systems as well as the 16-bit operating systems, like DOS, and you'll get significantly better performance than running on 32-bit chips."

Besides performance, another advantage Intel gains by moving to EPIC is ownership and control of its technology, something they lost to AMD and other competitors in the 1990's in court battles over the x86 architecture.

"With a unique and new instruction set, Intel sort of pushes the reset button on defensibility of the instruction set as intellectual property," says Bert McComas, founder and principal analyst of Inquest Market Research. "That doesn't mean it's 100% defensible, but who wants to test it in court against Intel?" This radical departure from the past is "a natural deterrent against anyone following Intel on that architecture," he adds.

Because AMD's Hammer is merely a superset of the existing 32-bit x86 chips, other chipmakers could follow in AMD's footsteps, potentially creating a competitive market for x86-64 chips. Transmeta Corp. is the first such chipmaker to license the x86-64 architecture from AMD.

Just as competition for 32-bit x86 CPUs brought down PC prices, AMD is hoping the same will happen with x86-64. To begin with, AMD is expecting Hammer to be cheaper to manufacture than Itanium because the Hammer CPU has only one-eighth as many general-purpose registers as the Itanium.

Registers are costly to implement, but they boost performance. Hammer's low cost and backward compatibility could make it more attractive for consumers and workstation applications, but its high-end computing performance will suffer compared with Itanium and other competing 64-bit computing architectures.

But Hammer could still find willing customers in the back-office server applications market that benefit from a 64-bit address space yet aren't concerned with raw number crunching.

"AMD will find a welcome customer base among servers and applications that are looking for an easy port to the Hammer's 64-bit large memory model," McComas says. "The most obvious example is big databases. What data-center manager doesn't want to have a huge amount of RAM and actually be able to use that all at once to solve database application performance bottlenecks?"

For customers to take advantage of the 64-bit address space, AMD needs 64-bit operating systems ported to Hammer. It's too early to tell if that will happen, but not all operating system vendors are enthusiastic.

"I'm not aware of any plans to ship a version of Red Hat Linux for the AMD x86-64 architecture," says Michael Tiemann, chief technology officer for open-source company Red Hat Inc. While Red Hat is actively pursuing Intel's Itanium (a 64-bit version of Red Hat Linux 7.1 for Itanium should be available now), it isn't committing to Hammer.

Meanwhile, Red Hat competitor SuSe Inc., makers of SuSe Linux, the Linux market-share leader in Europe, is driving the Linux port to Hammer, and that port will be available to the entire Linux community.

But on other mainstream fronts, neither Sun Microsystems nor Microsoft has immediate plans to support Hammer. "Sun has made the strategic decision to focus on Sparc and Solaris as our only platform," says Chris Kurell, Sun's computer systems marketing manager. Microsoft has confirmed that its existing 32-bit Windows operating systems should run on Hammer but hasn't committed to any 64-bit ports.

While AMD seems to be offering the right stuff-backward compatibility and a 64-bit address space-vendors aren't biting yet. The reasons are timing and competition from Intel. "It's still too far in advance of shipment, and the relationship between Intel and server vendors, unlike in the desktop and mobile space, is very good," says Rob Enderle, a research fellow with Giga Information Group.

"The AMD processor is actually better for the market today; however, strategically, Intel's is the better direction," Enderle adds. "Regardless of AMD's technology, its progress in this segment will be incredibly difficult given the conservative buyers and the fact that the Itanium now actually exists."