Improvements to Opteron are especially important now, since Intel recently neutralized many of AMD's distinguishing characteristics by adding them to its own server lineup. The most significant is 64-bit compatibility. However, faster bus speeds, larger cache repositories, power management features, the ol' hardware virus protection bit, and an advanced platform are also notable changes keeping AMD at the technological fore in this field.

The Latest and Greatest

Unfortunately, deciphering Intel's Xeon lineup is a somewhat complicated matter. You have multiple front side bus speeds, cache sizes, frequencies, and feature sets depending on which processors you're eying. AMD's Opteron lineup is much more straightforward, though it's starting to demonstrate some of the same variance as the company transitions from 130 nm manufacturing to 90 nm.

Up until recently, AMD has only tweaked one setting to differentiate its various Opteron models: frequency. An Opteron 250 was faster than a 248, which outperformed a 246 and so on. Every processor had a 1MB of L2 cache, operated across an 800 MHz HyperTransport bus, and plugged in to a 940-pin socket interface. In fact, most of those specifications persist today, but with a shift to 90 nm SOI manufacturing, AMD has added higher clock speeds, faster HyperTransport connections, and some new features.

In some ways, the changes look like a reaction to Intel's drastic Xeon revamping. Representatives at AMD counter that the revised Opteron core is merely a means to keep its architecture purring. When asked why accelerating HyperTransport to 1 GHz doesn't significantly bolster performance, Pat Patla, AMD's director of server and workstation marketing, touts the efficiency of existing platforms. "If there were a bottleneck previously, the move to 1 GHz would have really improved performance. As it stands, we enabled the faster HyperTransport link to prevent bottlenecks in future products."

The same line of thinking went into the addition of SSE3 instruction support. "In adding SSE3, we simply kept the processor competitive with other solutions. There aren't many server applications available with those specific optimizations, but as software developers continue working with SSE3, we'll probably see workstation-level programs better utilize the feature."

Smaller Size, More Efficiency: At 90 nm, the Opteron processor runs cooler, consumes less power, and costs less to manufacture than older 130 nm designs

Advanced power management is another one of those features to find its way into Opteron after Intel made a similar move. Make no mistake about it, though. AMD's Athlon 64 was the first desktop chip to incorporate power-saving features previously reserved for notebooks. PowerNow! is very similar to the more mainstream technology, Cool'n'Quiet, which dynamically reduces clock speed and voltage input according to load. The end result is lower electricity bills and diminished cooling requirements, both significant characteristics in blade servers and datacenters. Specifically, AMD's own measurements indicate a 75% reduction in power consumption at idle with PowerNow! enabled. Even under moderate duress, there's a 33% savings available.

If you're wondering why AMD waited so long to get power management functionality incorporated, it actually didn't. Opteron processors shipped after May 2004 innately support the feature. It's just not turned on yet. You'll need an enabling motherboard BIOS to get it up and running along with a potential operating system update. Did you sell any servers since last May? Now would be a good time to contact your customers and break the good news. There's nothing like a money-saving upgrade that doesn't cost a dime.

Pick of the Litter

Now some Opterons support PowerNow! while others don't. Some do SSE3, others don't. You get the picture. How do you make sure your customer gets the right chip? Fortunately, AMD's naming conventions eliminate the guesswork to some degree. If it's a single-processor workstation you're after, the Opteron 100-series is the automatic best choice. Dual-processor workstations and servers are enabled by the 200-series, while quad- and eight-way servers accept Opteron 800-series chips exclusively.

Plenty of Smarts: Much of the Opteron is L2 cache. However, an integrated memory controller and HyperTransport technology also give the chip its legs.

Interestingly enough, AMD is manufacturing all three of its Opteron families using the new E4 stepping. Both 200- and 800-series chips are available between 1.6 and 2.6 GHz are manufactured at 90 nm. The 100-series variants should emerge toward the end of April 2005. Though the differences are quite subtle, it's worth tracking down the newer chips for their power features and faster buses.

Expect the older processors to eventually disappear. AMD wants the new features included with all of its Opteron chips, not just the high-end models. It's only a matter of time before existing inventory is flushed completely.

New Chipset Components

Transitioning to a new HyperTransport bus speed is giving AMD occasion to release an updated chipset component, as well: the 8132 PCI-X 2.0 tunnel. The previous PCI-X device, 8131, featured two sides­—one that was 16-bits wide and an 8-bit side. Moreover, it supported transfer rates of up to 800 MHz.

Not only does AMD's 8132 boast two 16-bit sides, yielding more throughput to connected devices, but it also runs them at 1 GHz. Mode 2 operation enables PCI-X transfer rates of 266 and 200 MHz, both of which were unavailable with the 8131 PCI-X tunnel. According to AMD's launch documentation, its 8132 is specifically designed for use with PCI Express chipsets, too. In other words, you can expect to see it used with NVIDIA nForce Professional server boards.

A Bright, Shiny Future

If your customers keep up with technical happenings, there is a chance you'll have to field some questions about dual-core processing. In the very near future, both AMD and Intel plan to introduce products that incorporate two processing cores on a singular chip. When you install one dual-core component, Windows will recognize two processors, sort of like Hyper-Threading for Intel's Pentium 4. The principal difference is that Hyper-Threading operates on multiple threads while sharing the execution resources of a single core. Dual-core gives each thread its own resource pool, improving efficiency.

AMD is already claiming that any existing motherboard with 90 nm Opteron processor support will accommodate one of these dual-core chips from an electrical and thermal perspective. "We've been planning for this since 1999," says AMD's Pat Patla. "Everything has been designed to transition smoothly. Customers will need a motherboard BIOS update to recognize the processor, but dual-core is otherwise a drop-in solution." And just like that, single-processor workstations become dual-processor powerhouses, while dual-chip servers are able to execute four threads simultaneously.

You can understand how much this means to the server and workstation markets. A relatively straightforward upgrade promises to multiply the processing capacity of existing platforms, extending their useful lives far beyond what your customers would have imagined a year ago. As a reseller, you're in the position to go in two years from now with one upgrade that might add another two years of solid functionality.

Boxed to Move: When it comes time to sell dual-core later in the year, plan on offering boxed product to customers already running Opteron systems for a quick upgrade.

The real kicker here is that competing products won't make such a painless jump. Intel claims to be developing an entirely new platform to support its own dual-core contender, purportedly better able to cope with the throughput requirements of multiple cores on a single die. If you'll remember from stories past, the Xeon runs on a shared bus, meaning every connected CPU contends for a set amount of bandwidth. AMD's design is point-to-point, allowing each processor dedicated throughput and aggregate memory bandwidth. Consequentially, customers who've recently adopted the latest Xeon processors will be disappointed to learn Intel's newest technology will require a motherboard and processor upgrade down the road.

In Retrospect

AMD isn't making inroads exclusively with the small and medium businesses willing to try something new. Mike Rosenstein and his team have successfully lobbied several larger organizations to adopt Opteron, a point that you might want to mention when customers question the real-world benefits of a younger architecture.

Bell Helicopter, for example, recently transitioned its engineering and design teams to Opteron workstations. Computer Associates did the same. Even Microsoft Treasury made the switch in order to address massive financial datasets. There's really no shortage of corroborating evidence on AMD's Web site to support your pitch should a customer express apprehension about unproven technology.

Dependability, first of all, is a focal point in platform design and deployment. The chip is also flexible, meaning it's highly scalable from single-processor workstations to eight-way servers. Thirdly, Opteron enables backwards compatibility. Customers have the freedom to use their 32-bit applications today and adopt 64-bit down the road as software titles are ported. Though formerly a unique advantage to AMD, all of the 64-bit development until now has been with Opteron and Athlon 64, suggesting fewer transitional teething pains on those platforms.

And finally, there's performance, one of AMD's favorite subjects. Expect to hear more, as this spring will see an explosion of 64-bit operating systems, dual-core processors, and PCI Express peripherals. It's what AMD's Pat Patla likes to call the perfect storm of technology. Brace yourselves, folks.