AMD vs. Intel: The Next Phase

For those keeping a scorecard, AMD is the scrappy semiconductor vendor that's scoring big with its 64-bit Opteron server processor. Among many systems builders, the Opteron is considered a lower-cost and higher-performing option than Intel's competing 64-bit Xeon.

In 2006, AMD is poised to see its AMD64 architecture surge on the desktop. That's when Microsoft is due to ship a 64-bit-enabled version of Windows XP. Availability of the operating system will give a big boost to AMD's Athlon 64 desktop processor. (On servers, where Opteron plays, the lack of a 64-bit Windows hasn't hindered deployment, since Linux is a popular option.) Another indication of Athlon 64's growing mindshare came earlier this week, when it was named the editor's choice selection in the VARBusiness' 2004 Technology Innovator Awards.

While AMD is flying high at the moment, let's remember that the hare in the processor race is Intel. And there's a back story here, which AMD is very conscious of as it attempts to consolidate its gains. Namely, there have been times in the past when AMD has seemed like it was about to make a move on Intel, only to step on its own strategic toes and fall backwards

This time around, AMD isn't going to let that happen, vows Ben Williams, vice president of the company's server and workstation business unit. "If you look at AMD today vs. years ago, it's a very different company," he says. "Over the past few years, we have consistently executed our strategy."

id
unit-1659132512259
type
Sponsored post

He points to AMD's cutting-edge fab in Dresden, Germany, which is crucial to the company's efforts to ship processors made using new 90-nm technology. Many Athlon 64 parts are already available in 90 nm; Opteron is due to move to 90 nm soon. Dresden (actually, there are two fabs located side by side) will also be using the new 300-mm (12-inch) wafers, which are important because they yield many more chips than today's 9-inch wafers. Dresden will also be at the forefront of AMD's plans to ship multicore processors, beginning in mid-2005.

But Intel is also shipping 90-nm processors, and is planning multicore CPUs, too. Still, there seems to be a macro-level difference in the issues facing the two companies. AMD's challenges are ones of execution. Specifically, AMD has to prove it can continue to ramp up its production as it moves full-bore to 90 nm and eventually to 65 nm. It's also currently moving to 12-inch wafers. Dresden is so crucial to AMD's future because the company can't gain market share if it can't ship more parts. Williams points out that Dresden has more than ample manufacturing headroom to satisfy market demands.

Intel's problems, on the other hand, seem to be more strategic in nature. The company is renowned for its fab smarts, and rightly so. But lately it seems like Intel has made some poor microprocessor-design choices. The latest case in point occurred earlier this month, when Intel quietly began letting its customers know that it pulled the plug on a planned 4.0-GHz speed grade of its Pentium 4 processor.

"This is a customer communication," an Intel spokesman told me, adding that the company wasn't planning to issue a press release.

Word around the industry was that the 4.0-GHz part was scuttled because it would have dissipated too much power. (That's something Intel won't have to worry about with its coming multicore CPUs, which will avoid any power problems, at least for the next few years.)

Intel took a publicity black eye because the news marked the second time in recent months that it has changed its P4 plans. In July, Intel announced that the processor, which had originally been set to ship in the fourth quarter of this year, was being pushed back to the first quarter of 2005. Intel said the reason was that it needed more time to ramp up its 90-nm production.

The P4 delay was doubly problematic for Intel because it followed a string of glitches over the previous six months, including the discovery of defects in its Grantsdale P4 chipset, the scrapping of the planned Tejas processor because it ran too hot and delays to the Dothan Pentium M mobile processor.

Still, Intel seems to be recovering ably. Remember that the P4 family itself isn't going anywhere. Rather than painting the axing of the 4.0-GHz part as another problem, Intel is positioning the move as a way to focus the company's processor road map on factors other than clock speed.

"I think we're really focused on end-user benefits, such as ease of use, security and hyperthreading," the Intel spokesman added.

Most immediately, Intel says it will bring larger 2-MB caches to its entire 90-nm P4 line during 2005. While critics point out that big-cache parts are more expensive to make because fewer of them fit on a wafer, advocates note CPUs with large caches can perform as well as smaller-cache processors with slower clock speeds.

At least one systems builder isn't worried about where Intel is headed. "Intel has the CPU, chipset and the motherboard," says Robert Schaffer, president of Randolph, N.J.-based Source Micro. "From the platform side, the stability and the warranty support are impossible to match." In contrast, AMD is a CPU-centric supplier.

Still, as Schaffer notes, "Intel has made a whole load of changes very quickly." He points to the shift from Socket 478 parts to the new Socket 775, which hosts newer P4 parts and is teamed with the new Grantsdale chipset.

Schaffer says that the move to Socket 775 seems to have resulted in some production shifts at Intel, which have put a bit of a squeeze on the availability of older 478 devices. That has caused VARs with unfilled customer orders for 478 machines to have to do a bit of a dance to obtain the parts they need.

However, the situation should ease as the transition to the new socket progresses. Also factored out of the equation -- because the playing field is level for all VARs -- are two recent processor pricing changes by Intel. This summer, it altered the list prices of many of its desktop and server CPUs, making up for the cancellation of a boxed-processor rebate program by instituting broad cuts. This month, Intel did the same for its mobile devices. "The pricing has become a non-issue," Schaffer says.

Another burning issue was also put to rest this week, as Microsoft weighed in on how it will charge for its operating systems running on the multicore parts both Intel and AMD are planning. Some observers were worried Redmond might get greedy, trying to charge for each core (e.g., two licenses for a dual-core CPU) and thus angering customers. That won't happen, as this week Microsoft apparently decided that maintaining market share against Linux is the better part of valor. Accordingly, it will charge by the socket, not the core. So, a dual-core CPU will require only a single Windows license.

As this recent raft of moves from AMD and Intel filter down to the channel in the coming months, the big question about the race between the two remains unresolved: Will it be the tortoise, or the hare?