Analysis: Will AMD’s Opteron chip become a high-end player?
AMD Corp. this week officially introduced its 64-bit Opteron chip. Analysts say that Opteron boasts an advantage over competitive chips from Intel Corp. because it can run both 32- and 64-bit code without sacrificing performance.
But history shows that AMD was also ahead of the curve when it introduced its Athlon chip in 1999. Athlon boasted faster clockspeeds, a more efficient bus and an L2 cache design that was similar to that found on Intel’s high-end Xeon processors -- but failed to make a dent in the midrange or high-end server markets, largely because of a lack of OEM support. This time around, representatives from both Sun and IBM have expressed interest in Opteron. But is either vendor up for the challenge of developing a scalable multiprocessor chipset to support Opteron on its high-end systems?
With impressive performance in crunching both 32-bit and 64-bit code, AMD hopes that Opteron will challenge high-end processors from Intel. Provided that AMD can snare some high-end partners, that is.
Opteron's ability to execute 32-bit and 64-bit code without sacrificing performance might seem like a no-brainer, but for the fact that Intel’s flagship 64-bit Itanium microprocessor is considerably slower than its 32-bit Xeon chips when crunching 32-bit code, which it has to run in an emulation layer.
A list of OEMs -- including server blade specialist Racksaver Inc., chipmaker Newisys Inc. and graphics giant Nvidia Corp. -- have announced initiatives timed to coincide with AMD’s Opteron launch.
Racksaver, for example, will ship a four-way server -- the QuatreX-64 -- based on Opteron, while Newisys and Nvidia are prepping mainboard chipsets for the new processor. Moreover, AMD has also said that at least two major OEMs are interested in selling Opteron-powered servers. Some industry-watchers -- such as Gordon Haff, an analyst with consultancy Illuminata Inc. -- believe that IBM Corp. and Sun Microsystems Inc. will eventually ship Opteron-based products.
In addition, a number of ISVs -- Microsoft Corp., Oracle Corp. and Computer Associates Int’l Inc. among them -- have announced support for Opteron-specific extensions in their products. Microsoft has committed to supporting Opteron in versions of Windows 2003 Server and Windows XP, while Oracle and CA have announced Opteron support for their respective database products. Microsoft currently supports Itanium in a special version of its Windows 2000 Advanced Server product -- Windows 2000 Advanced Server Limited Edition -- as well as in a 64-bit version of its SQL Server database. Specifics about the software giant’s planned Opteron support among its Windows Server System family of applications have up to this point been vague.
Moreover, other big players, such as Dell Computer Corp. -- which doesn’t currently market an Itanium-based server -- and Hewlett-Packard Co. (HP), have been much less vocal about their support for Opteron. HP, in particular, has been heavily involved with Itanium development since the mid- to late-1990’s.
As Opteron makes its debut, most of the emphasis -- at least, on the part of both AMD and its partners -- seems to be on the low-end. Newisys, for example, will market single, dual and quad-processor chipsets for Opteron. Nvidia’s NForce 3 Professional chipset, on the other hand, is designed for engineering or professional workstations.
That’s in stark contrast to the Itanium space, in which SGI currently ships a 64-way Itanium-powered supercomputer -- the Altix 3000-- and HP has announced a 64-way Itanium-based Superdome system, while vendors such as Unisys Corp., and NEC have introduced 32-way Itanium-based servers. Several months ago, Fujitsu Ltd. and Intel even made noises about developing a 128-way Itanium-based mainframe. At this writing, HP and Unisys have also disclosed roadmaps for 128-way Itanium-power systems.
The requirements of large SMP systems are pretty consistent across all chip architectures, RISC, CISC and EPIC alike: High memory bandwidth, low-latency and fast I/O performance.
In this respect, the vendors that are marketing large Itanium-based systems have tackled these issues by introducing proprietary chipsets designed to support a large complement of processors.
An instructive history lesson
AMD faced a similar problem when it launched its cutting-edge Athlon microprocessor in August 1999. In Athlon, AMD had a chip that clock-for-clock outperformed Intel’s then-dominant Pentium III microprocessor. And for midrange server applications, Athlon exploited an Alpha EV6 bus that was originally developed by Digital Equipment Corp. This gave it plenty of bandwidth -- up to 1.6 GB/s -- to support large SMP configurations of up to 14 processors, analysts said at the time. And also like Intel’s then-flagship Xeon chips, Athlon also featured an L2 cache that ran at the same speed as the microprocessor itself. Athlon, it seemed at the time, was an all-purpose chip ideal for both desktop and midrange server applications.
There were a couple of catches, however: Support from major OEMs was lacking, of course, and AMD also hadn’t demonstrated a scalable chipset that would support large SMP Athlon systems. At the time, AMD expressed optimism that major OEMs would support Athlon in their systems, and pointed to the efforts of third-party vendors such as Hotrail Inc. and Alpha Processor Inc. to develop a scalable SMP chipset for the next-generation processor.
The result was something of a mixed bag. Major OEMs did eventually introduce support for Athlon -- on their desktop systems. And although a small Athlon server market does exists, bolstered by the release of an Athlon multiprocessor chipset in 2001, it didn’t cut significantly into Intel’s share.
That being said, industry-watchers suggest that there are a couple of reasons why things could be different this time around. “There are some faster 32-bit processors, in particular Intel Pentium IV Xeons running at 2.5 GHz,” said Nathan Brookwood, a principal with microprocessor consultancy InSight64. “For organizations with a substantial investment in 32-bit applications and with no pressing need for 64-bit support," Brookwood agreed, “there is not a lot of incentive to move to [Itanium] right now. This could create an opening for Hammer [AMD’s code-name for Opteron], which runs 32-bit code without sacrificing performance, when it appears.”
Vendor support key
Of course, Opteron will only become a viable platform for midrange and high-end customers when it is supported in large SMP systems by major vendors.
For a variety of reasons, Analysts such as Haff and Brookwood expect that both IBM and Sun will introduce Opteron servers. Representatives from both vendors have indicated that their companies will at least consider developing systems based on the chip. The only question, it seems, is what role Opteron will play in their respective product lines.
Take Sun, for example. Although the Unix giant initially indicated that it would deliver a version of its Solaris operating environment optimized for Itanium, it later backed off of that promise. Things came to a head, last August, when Sun chief Scott McNealy bashed the Intel chip, dismissing it as “Itanic.”
Since then, Sun representatives have quietly suggested that Opteron -- and not Itanium -- may play a role in future systems, similar to its LX-50, that are designed to run Linux and Solaris x86.
This week, Sun issued a statement in which it promised to support Solaris x86 -- which is available only in 32-bit binaries -- Java and the Sun ONE software stack on Opteron. Sun pointedly didn’t address the issue of 64-bit support for Opteron in either Solaris x86 or the Sun ONE software stack, and the Unix giant made no mention of whether or not Opteron support will have a place in its Orion initiative, which consolidates all of its software offerings in a single subscription package. It also hasn’t yet said anything about developing Opteron-powered systems, either.
IBM, for its part, does support Itanium in its xSeries line of Intel-based servers. But to this point, Big Blue hasn’t been as gung-ho about Itanium as HP, largely because it has no intention of replacing its own G-series mainframe and Power-series RISC processors with Intel’s upstart 64-bit chip. In an interview last month, Deepak Advani, IBM’s VP of high performance Intel servers, stressed that IBM has a substantial commitment to Itanium, but suggested that Big Blue could also give Opteron a look when it becomes available. “Unlike some of our competitors [Sun and HP] we’re not overly-committed to the success of a single architecture, so if it makes sense for our customers that we [introduce Opteron-powered servers], it’s something that we might consider doing.”
This week, IBM indicated that it would support Opteron in a future budget-oriented server product.
With a big player like IBM onboard, Opteron could really fly. After all, Big Blue developed its own “Summit” chipset for large Intel NUMA-systems, introducing a 16-way x440 server in December. IBM has also announced plans to ship a 32-way Intel-based server before the year is out. Big Blue has a great deal of experience in designing large SMP systems, as well: It currently supports 32 Power4 processors in its top-of-the-line p690 AIX server, and its next-generation mainframe -- code-named “T-Rex” -- is expected to eventually support as many as 64 G8 processors in an SMP configuration.