In the second quarter, AMD held a 26% share of the x86 server processor market, according to Mercury Research.
Randy Allen, corporate VP of the server and workstation division of AMD, led the original design team for Opteron at AMD beginning in the late 1990s. He talked with InformationWeek about the expected impact of the new Opteron.
InformationWeek: Opteron Rev-F is not a major new architecture change like we saw from Intel recently, but more in the vein of adding new bells and whistles to the existing architecture. How big a change are we really looking at?
Allen: One of the things we've heard from our customers quite a bit is their desire to not have a lot of disruption in the underlying architecture. In fact, we're excited about the performance levels that are going to be delivered by this product without changing the system architecture in a dramatic way.
Obviously over time there are necessary disruptions to any platform. This represents just our second socket for the Opteron. This transition was really driven by the change in the industry from DDR (double data rate) to DDR2 memory and not just a whim from AMD to force a transition on our customers.
InformationWeek: Intel has been using DDR2 memory for some time. Is this a case of AMD being slow to jump on new technology?
Allen: We've believed DDR2 was the right technology, but it is just now getting to where it can be cost effective in delivering real value to the end user. This isn't a race to see who can get to some technology point fastest. Technologies should be introduced when they can provide real value.
InformationWeek: What do you believe are the most important advances of the Opteron Rev-F?
Allen: It is really the entire package. Obviously we view the enhancement in enabling the DDR2 memory technology is probably one of the strongest contributions to the next generation processor because it does provide for higher levels of performance and performance per watt. Most important longer-term is the quad-core upgradeability. Each of these platforms using (Opteron Rev-F) will be ready to just drop in a quad-core processor and provide unprecedented performance levels in mid-2007. And one of the things that is clear to pretty much all IT executives is that virtualization is going to be the next big thing in managing all these large data centers. Some of the new capabilities we are providing in that space are going to make things easier on both the software ecosystem in terms of their solutions for virtualization, but also are going to directly deliver higher performance levels for these virtualized environments.
InformationWeek: Last week all the major virtualization software vendors announced plans to support your embedded virtualization technology. How quickly are they going to be able to bring anything significantly new to the market?
Allen: Virtualization is going to be an area that continues to receive innovation both at the processor level from companies like AMD, and the hypervisor level from folks like Microsoft, ZenSource, and VMware, and certainly at the operating system level as well.
There is not going to be a virtualization solution that is signed, sealed, and delivered, and that's it. We have a roadmap based on conversations we've had with all our partners where they actually specified the features for the processor introduced this week. We've also engaged with them on features that will be folded into our longer term roadmap.
A lot of the benefits will be delivered in the coming weeks and months. We are seeing that a lot of end users are taking virtualized environments out of the lab and putting them into production at a fairly rapid clip. InformationWeek: Are the server manufacturers using Opteron Rev-F to simply upgrade the product lines they already have in place, or will Rev-F expand the number of total systems available in the market based on Opteron?
Allen: That is one of the things that we are most excited about on this launch. The number of Opteron platforms doubled from 2004 to 2005, and it's going to double again in 2006. We are going to see much broader and deeper product offerings from companies like IBM. This is not just a refresh of existing platforms.
That makes sense. Many of the [system manufacturers] that don't have broad Opteron portfolios have watched as Opteron has really taken off in the market. This is an appropriate time for them to target their development efforts in conjunction with this next-generation Opteron.
InformationWeek: Some early benchmarks indicate that Intel's new Woodcrest server processors are much more competitive with Opteron now on the market than Intel's prior line of Xeon processors. How important is it to demonstrate clear leadership over the next few months?
Allen: I think it's very important for us to make sure people are aware of the superior performance and performance per watt of this next-generation Opteron.
There are several benchmarks where Opteron outperforms Woodcrest in the two-socket space by a significant margin. It is also true there will be workloads out there where Woodcrest outperforms Opteron. It is going to be very dependent on the workloads an end user is running to determine who has the performance crown.
In the four-socket space that is not true at all. Since we introduced Opteron we have been running circles around the competition in the four-socket space. It is true they have a much more competitive product in the two-socket space, but in the four socket space the margin has not been reduced by an significant degree.