Advanced Micro Devices’ Dirk Meyer has presided over many chip development teams. But as senior vice-president of AMD’s Computation Products Group, lately his job has meant more time spent in front of airport security screeners than processor designers as he travels around evangelising AMD’s eighth-generation Opteron and Athlon 64 processors. One year after Opteron’s introduction, Dirk Meyer was talking it up for a group of financial services customers. Whilst there, he took some time while to talk to Tom Krazit about the progress of Opteron from the breathless hype of the pre-launch marketing to the steady progress the chip has made with server manufacturers and customers in the enterprise world over the past year.
What do you need to do to make sure Opteron builds upon the toehold it has taken in the enterprise this year?
Dirk Meyer (DM): Much of the initial adoption was in the high-performance technical computing clusters, and as we look forward, we aim to continue to be strong in that space, but also grow across the breadth of the enterprise. Interestingly, what’s required there isn’t so much more of a different technology beyond that which we are delivering today, but rather the right sorts of relationships with original equipment manufacturers (OEMs) who will put Opteron in enterprise class boxes.
We’re fortunate in that we now have those sorts of relationships, and you’ll be seeing some of those products out in the market. In fact, HP is out there now with enterprise-class (four-processor) systems. So, the near-term objective is to broaden our relationship with our tier one OEM customers to get the Opteron platform available (in servers) that are more suited to the general needs of the enterprise.
The next step of course, is the software ecosystem around AMD64.
How important is Intel’s decision to enter the market in building out that ecosystem?
DM: I think it’s going to accelerate the rate at which AMD64’s ecosystem gets developed. Now that Intel is there with the same instruction set, essentially, the business case is even more clear to the independent software vendors (ISVs). I think that decision by Intel only serves to accelerate adoption of AMD64.
How many customers are currently running 64-bit applications on their Opteron boxes?
DM: Actually, in the high performance computing (HPC) space, on top of Linux, that’s where we see it mostly today. In the enterprise, the pickup is a little slower, largely because they have such long certification cycles. We see 64-bit Linux out there that people are playing with, and some of the databases people are looking at.
Do you think the delay in the Windows availability has retarded that development a little bit?
DM: Well, it certainly doesn’t help. But honestly, the movement toward 64-bit pervasively across the enterprise isn’t something we expected to happen over the course of small single-digit quarters. These big enterprises are pretty conservative; they have long certification and qualification cycles. A lot of them have already been playing for quite some time with the Windows operating system from a beta perspective. So yeah, that’s slowed it down a little bit, but I wouldn’t say fatally or fundamentally.
Is the availability of a 64-bit Windows operating system a key buying decision with some of the potential customers out there? Are people delaying their purchases until that arrives?
DM: I was talking to customers from financial institutions, and they all understand the value proposition Opteron has as a 32-bit machine. And the adoption of Opteron in the enterprise today has been more limited by the availability of enterprise class systems from OEMs, which of course are now starting to show up from Sun and HP. Next, these guys need to get the boxes out in their environments; qualified and running the applications they are running today. Opteron delivers such phenomenal performance in today’s 32-bit environment that there is huge demand for those boxes today, independent of 64-bits. So from what I’m seeing, the 64-bit play is really the icing on the cake.
From an architectural standpoint, is the chip as it is today something that will be able to work on a majority of enterprise workloads?
DM: First of all, it’s interesting to note from the server volume perspective that there’s a huge volume inside the enterprise represented by (two-processor) and (four-processor) servers. In particular, the (two-processor) servers are often the 2U (3.5-inch) boxes with redundant power supplies. And that’s a configuration that we don’t have on the marketplace yet.
A lot of the work we need to do in the near term is to broaden our product mix with the OEMs to get enterprise-class boxes out there.
When you say near term, are you talking about the next year?
DM: I mean over the next 12 months the focus will be to work with OEMs to get those class of machines into the marketplace and also to get work on our partners, our ISV partners, to get 64-bit solutions out into the market, and finally to continue to evangelise the technology with end users in the enterprise.
Over the next 12 months, we’ll of course have more speeds and feeds on processors, but our focus is more on the customer relationships and ecosystem development.
As you look toward the end of the decade, will AMD focus on getting Opteron into large symmetric multiprocessing (SMP) servers?
DM: In the near term, that’s a non-focus area. There’s so much volume in the server space that we can grow into in the kind of configurations I outlined earlier. Clearly, over time as we get deeper and deeper into the enterprise and more and more important to the OEMs with whom we’re partnering, those sort of bigger SMP systems are going to be of interest. But generally, those are the sorts of things we would only engineer in pretty close partnership with the big OEMs.
And of course, the tensions there right now that the global tier one (vendors), outside of Dell, which doesn’t really have big SMP systems, those that do have their own proprietary architectures. My view is that AMD64 is as well suited for big SMP machines as it is for scale-out environments, and it’s just a matter of time. But, of course, that’s something we’ll do in concert with our OEM partners.
It’s all about opportunity cost from an engineering side. Right now, as we look to broaden our portfolio across both servers and low-power mobile devices, clearly development in big scale-up machines isn’t a priority today. But as we grow and broaden I expect it will become more of an active priority for us.
Why do you think Dell has so far declined to jump in?
DM: Well, one thing to observe is that Dell’s not really a market maker, they’re a market taker. So we didn’t expect Dell to lead the charge, and they’re fulfilling our expectations.
Do you have any sense of when that market-taking opportunity for Dell might come?
DM: I think that as Opteron grows across the enterprise and starts taking business from them that they’ll respond.
What would have done differently over the last year, in regards to Opteron?
DM: Given how quickly the product is being adopted, with perfect hindsight, I would have chosen invest a little more aggressively than we have, relative to our server product line. But I’m pretty satisfied with the past year.