By William Van Winkle
Few would argue that AMD is a company of historic greatness, visionary innovation, and formidable nerve. When Opteron had Xeon on the ropes and was demolishing Intel’s kidneys, the Little Titan of Texas wasn’t shy about trumpeting its successes. “There’s no reason why AMD can’t achieve 40% [server processor market share],” proclaimed commercial business vice president Marty Seyer in the summer of 2006. Humility has never been one of AMD’s strong suits...perhaps until last December, when CEO Hector Ruiz admitted, “We blew it and we’re very humbled by it and we learned from it and we’re not going to do it again.”
Are you ready to believe him?
Let’s not make a catalog of the ills that befell AMD in 2007. What’s done is done, and with the volume re-launch of quad-core Opteron (Barcelona) based on B3 silicon in March, even the infamous TLB erratum bug is behind the company. Adversity breeds innovation. AMD has been under the gun before and come out a winner. The key question for channel resellers today is whether AMD is set to pull another rabbit out of its hat.
This cover story won’t take you into the technical nitty gritty of K10 architecture, unified shader design, or any of that ground that AMD has scaled in the last year. We’ll touch on a few key points, but by and large we’ve delved into these points in past issues. Right now, it’s time to talk about the company and its significance for the channel. This is about being smart and picking the right partners. We’ve seen a lot of problem-plagued products and executions in the last year, but does this mean that 2008 will be a repeat of 2007? The early indicators would say no, but conjecture will only take us so far.
If you were fortunate enough get in on beta testing Windows Server 2008, you probably have a good idea of what the operating system can do and how it improves Server 2003. So much has changed, though, that it would be difficult to put each and every new feature to the test. And if you haven’t had the chance to go hands-on with Server 2008 yet, it’s still a blank slate of opportunity waiting for your mark.”
We pursued some of the key channel people at AMD to ask the hard questions and get real answers. What you’ll see in the following pages are the most important exchanges from hours of conversation with the executives most directly tied to AMD’s channel efforts. Are their comments and answers biased? Of course. Just keep in mind that AMD remains keenly aware that it is on a sort of probation with many resellers. Bridges to the channel may have been charred in 2007; this will be the year when those bridges are either repaired or burned out altogether. Rose-colored glasses can make any company look good, but AMD’s long-term interests might benefit most now from brutal honesty. That’s what we’re after.
Whether servers are already part of your offering mix or you’re mostly new to the space, this is a rapidly expanding product category rife with opportunity and applicable to businesses of all sizes, even fledgling garage operations. However, you need to understand the options available within the SMB server category before you can recommend the right specific configuration for your client and ace out the competition. So from entry-level to volume racks, we’re going to breeze through the field and make sure you’re up to speed on the essential hardware options.
AMD ON GRAPHICS:
RAM: The TLB issue squashed a lot of the momentum and messaging AMD was plowing into the Spider platform. The 780G chipset was a key piece of that platform, and it’s been racking up a lot of rave reviews that may have passed under the radar of some resellers. Is this a game-changer in the desktop space? It seems to have opened a wide feature gap versus mainstream Intel chips.
JB: Absolutely. Intel’s a worthy competitor—that’s the understatement of the century. But we’re really not worried about it. We’ve got true PCIe 2.0. We’ve got a great overclocking facilitator called Overdrive. We’re supporting DisplayPort from top to bottom now, and we’re the only company today able to claim that. We’re supporting hybrid graphics, unbelievably low power, and also it’s quiet as hell. It’s just a great, great product.
RAM: Will the 780G impact low-end discrete graphics sales?
JB: For sure, it’s eating into the entry level of the discrete graphics segment. We had the 2400s in discrete and now the 3400s. If you look at the worldwide GPU market, we’re seeing TAMs [total addressable markets] being relatively flat, but we’re seeing ASPs [average selling prices] pick up dramatically. For sure, 780G has raised the bar on IGPs as well as where value-add discrete graphics start in the market. This is not something you’re going to find on a $35 motherboard, but there’s still huge demand for it.
RAM: What are hybrid graphics?
JB: In layman’s terms, it allows the 780G chipset to be able to communicate with the 3450 and 3470 discrete GPU, combine the processing power of both cores, and give you performance scalability to 1.8X. In the past, if you had integrated graphics and went and bought a graphics card, then normally the graphics [capabilities] in the chipset were disabled. Now you have a genuine opportunity for the two to talk to one another. That’s a great feature for us that Intel hasn’t been able to do because they don’t have discrete graphics cards. And we’ve all seen games giving you a blank screen when running with hybrid NVIDIA. This doesn’t. This is running Crysis, Call of Duty 2, Half-Life, DOOM 3—you name it, we’re running it. Look at the benchmarks against, for example, [Intel’s] G35. Crysis doesn’t even run in DX10 on that. Frankly, 780G rocks.
RAM: Is DisplayPort a big deal?
JB: It’s like asking if HDMI is a big deal. Three years ago, people would have said, “HD-what?” Now, you look at most people’s high-def TVs and they have HDMI. DisplayPort is absolutely a big deal. It’s the ultimate replacement for HDMI. In terms of the global channel, we believe you’re going to see a lot of DisplayPort cards coming into the market in Q4 of this year. By the second half of 2009, everyone’s going to be wanting DisplayPort.
RAM: Let’s talk a bit about the relevance of the ATI acquisition to resellers. Would the HD 2000 and HD 3000 products and launches have been the same without ATI’s joining with AMD?
JB: The 2000 was too far along in its cycle at that time, so yes, it would. The 3000? That’s a good question. Our engineers did a phenomenal job on the 3000. Would it have gone off as smoothly without AMD? I’d be loathe to say yes or no. I will say that this is the first time in my years with ATI that we brought products to market on A11 silicon. It was nothing short of remarkable. You want to talk about execution? There’s perfect execution.
RAM: In hindsight, was the ATI acquisition the most prudent move for both parties?
JB: I’ll try to give you a balanced perspective. Has it been a challenge in 2007? Of course. You’ve seen the results. However, we’re the only company right now that can genuinely deliver platforms to the marketplace. You’ve got NVIDIA claiming we no longer needs CPUs, and you’re seeing Intel say you no longer require a fast GPU. We’re saying, hey—we can offer both. We’re in a great position to support AMD on AMD, but you’re going to see the first real migration of the technology coming in Fusion, and you should see it shipping in the second half of 2009.
But the other thing is we’ve got over 8,000 engineers now. We’re dealing with every major OEM in the world. And from an ATI perspective, it gives us the ability to have a bigger sales team and a significantly larger marketing company. As an example, China has a huge cybercafe market. Now, with ATI, we had one person supporting all of the cybercafes in the China market. As part of the greater AMD, we’ve got about 200 people addressing that marketplace. In the short term, you look at the numbers and say, ‘Oh, my, maybe that wasn’t such a good thing.’ In the long term, we’re beginning to see the fruits of both of these companies coming together.
RAM: Many people think the main reason for the merger was that producing Fusion, or something like it, was essential to the viability of both companies.
JB: Fusion is going to be one hell of a product when it comes to the market. Integrating the GPU onto the CPU, allowing the end-user to have the ultimate visual experience, is an exciting thing. Intel had a CPU, and now they’re working as hard as they can to get Larrabee into the marketplace. So they see the requirement of strong 3D graphics. The reason NVIDIA has turned its guns on Intel is the company realizes they’re pretty exposed. It’s a lot like a stand-off at the OK Corral. One guy’s got a machine gun, and the other guy’s got a pistol.
RAM: Given recent history, should the channel give credence to Fusion’s current time line?
JB: 2007 was a painful lesson. Basically, there are only so many chances you can have before people give up on you. I believe now, with our engineers in place, with our systems put in place, with the two teams collaborating very closely . . . yeah, we’re comfortable with the second half of 2009 schedule.
RAM: With all the video work we do on RAM TV, the people at this magazine are big workstation proponents. But we haven’t heard much about professional ATI graphics in a while.
JB: That’s a 4.5 million-unit market—about 2.5 million 3D and 2 million 2D—with over $600 million in revenue and people working at gross margins of 60%. NVIDIA, with the Quadro, has been the dominant player in that marketplace, capturing between 80% to 90% of the market, depending on the quarter. We had a bit of a challenge in that our hardware wasn’t that great, and our software wasn’t much better. However, now that we have these Linux-certified drivers, now that we have some very, very strong hardware, we’re very confident about our technology. You talk about whether it was good for ATI and AMD to come together. We now have a development team here that we just did not have at ATI. These people are bringing opportunities to us from the industrial segment, telecom, education—opportunities we just wouldn’t have had at the old ATI. So we’re very excited about that marketplace—and frankly, it’s the heart of NVIDIA. That’s a huge cash cow for them, and we’re going for it.
RAM: Going for it how?
JB: Like everything else, it starts with products. If you look at our 8650 and 8600 coming to the marketplace, for the first time in as long as I can remember, people—reviewers—are saying that ATI has the leading-edge performance in professional graphics. The other thing we haven’t had is software. We haven’t had Linux driver support, and now we do. We’ll also be working with AMD’s classic CPU distributors to bring these parts to the worldwide channel, something we were never able to do in the past. The opportunity is huge. Even the 2D space for our FireMV line—that’s a $130 million market composed mostly of financial traders—and we think there’s 2D opportunity beyond that.
RAM: This brings up the perennial leapfrogging war with NVIDIA. Obviously, when the single-GPU HD 3870 didn’t take out NVIDIA at the top-end, it raised a lot of reviewers’ eyebrows. But does AMD still care about this leapfrogging game?
JB: First off, let me say that a lot of the reviews we’ve been seeing globally show that the 9800 GX2 wasn’t warmly received. Not at all. They had huge problems with the quad driver support. They had problems with heat dissipation. And with our X2, there are two graphics cores on one PCB. If you look at the NVIDIA boards, it’s actually one graphics core on a PCB and two PCBs bridged together. People just didn’t like it. Also, depending on the benchmarks you’re working on, the 3870 X2 is faster than the 9800 GX2. However, the 9800 GX2 is $499 to $549. The 3870 X2 is on the market at $399. There’s such a big price gap and minimal performance gain, we believe we’re in good shape.
Now, it used to be that if, say, the 3870 X2 was king of the hill, in every other segment below it you would win because you had that king-of-the-hill product. What we’re seeing now is that you almost have to be king of the hill in every segment. As an example, if you look at the old 2900 product, that did reasonably well in the reviews, but it also got panned in a number of publications. However, 2600s and 2400s were phenomenal. We sold a ton of them. In Q4, the 2600 was the No.1 selling product in the worldwide channel. Worldwide channel means 45 to 48 million PCs. If you look at the space where the 3850 and 3870 sit, namely $179 to $229, a TAM analysis a year ago would have showed that space was 250,000 units per month. We now realize that space has become 600,000 to 700,000 units per month, which is great because it means that the graphics message is being heard loud and clear by consumers. Ultimately, if we don’t have a good product in that $179 to $229 range, it doesn’t matter what happens to the 3870 X2. Yes, we care about competition at the high end, but we’re looking for clear leadership in every segment.
RAM: You’re fighting on multiple fronts, not just for each sale but also for public perception of ATI versus NVIDIA overall. How do you prove to the public and resellers that NVIDIA isn’t the last word in graphics?
JB: If I were a magician, I’d have resolved this already, right? I’ve been in this market a long time, and people will say that ATI’s a lovely company with nice products, but it always comes down to NVIDIA. So we’re having to work hard to change that perception. We have a global channel program called RAP, Radeon Authorized Partner. We’re capturing the top channel customers and finding a lot of their business when they buy from ATI. I mean, when they buy a product from one of our authorized add-in-board partners, we’ll fund them dollars. Previously, when we had this program, we had 500 customers and couldn’t leverage the dollars. Now we can. By receiving those dollars, those customers are actually providing us their bi-weekly sell-out data, the purchase data. So very quickly we can see the worldwide channel, what’s selling, what’s not, do we need to make pricing moves, do we need to improve some more? We’re also running a demand generation program with each of the top 200 customers.
Also, NVIDIA had a strategy called “The Way It’s Meant to Be Played.” The reality is that with the 2900 family of products, the 2900 was late. So if you look at the ISV community, they want hardware to code the games, and they want some investment. Because the 2900 was late, it gave NVIDIA a chance with the 8-series to really capture those ISVs. This year, we’re investing significantly more dollars to work with those top ISVs, investing more in terms of people, in terms of marketing programs. This is on top of everything AMD does with its regular channel reseller program.
RAM: That all sounds promising, but, as you say, it starts with products. What can we look forward to in your hardware?
JB: The official line is we don’t comment on our next products. However, I will say that we’ve got the next generation back in the labs, and they’re looking great. The timing will be in the summer. Again, we’re looking for leadership in each segment, and I think if you look at the RV670, meaning the 3850 and 3870, that got us back in the game. We were three months before the 9600 GT, which has been one of NVIDIA’s worst launches in almost two years. If you look at RV770, it gets the clear win. That’s what we’re shooting for, and it’s going to be great.
AMD ON OPTERON
RAM: Let’s set the record straight. What happened with the recent history of Opteron?
JF: We introduced the product last September, and about a week after that, we received notification from our engineering teams that they had identified an errata that would potentially have an impact on a very small number of people, but based on the fact that these are processors going into servers that could be running a lot of critical functions, we made the conservative choice to hold off on shipping so we could rev the silicon and take care of the issue. In the meantime, there’s been a lot of back-end work to change the silicon, do all the testing and validation to make sure everything was good before we release the product back into distribution.
RAM: What was AMD doing in the server market during this period? Everything seemed pretty quiet.
JF: AMD didn’t just go radio silent and say, ‘Look for us in six months!’ We were working with partners to identify the key Barcelona-related deals, figure out the customer needs on those, and deliver something in the meantime that allows them to not have to put their businesses on hold. So we provided some earlier revs of the silicon, in some cases dual-core processors with a very clear upgrade path to quad-core. We did a variety of different things so people could continue on these projects and continue to build out the infrastructure they needed.
RAM: Businesses wanting Barcelona were fine with deploying dual-core?
JF: The key was to minimize financial impact on the customer. We weren’t saying, ‘Hey, come buy our duals, and then you can buy the quads when they’re ready.’ We were making an investment in the business and giving them an aggressive price that would let them net out where they would have if they’d bought the quads to begin with.
RAM: Smart. What has the net effect been for AMD?
JF: We haven’t grown market share, but we also haven’t lost market share, and one would expect that if we were disadvantaged from a product perspective. There were two things that played into this. One was that our competitor’s product really wasn’t a great quad-core solution. It kicked out about 50% more heat and drew 50% more power than the duals and it didn’t scale very well. So it wasn’t a very viable solution for most customers. Customers saw that we were going to have a better platform and a better solution. You know, customers think about things in the long term. In the desktop world, the decisions are pretty short term when you’re revving your machine every year. But when you’re planning a data center with an architecture you expect to have for three to five years, you’re willing to wait a little longer to make the right choice. We started shipping Barcelona in March, but you probably won’t see parts on shelves until at least mid-April because we’re still so busy fulfilling all the backlogs from people who said they were willing to wait.
RAM: AMD share prices tell one story, but how much would you say that not having Barcelona hurt the company over the last several months?
JF: If you pull up IDC server shipments, the x86 server markets at the end of Q4 had us in a market where almost three-quarters of the market was still dual-core or single-core. I think quad-core gets a lot of attention because it’s the new, hot thing, but the reality is that the bulk of what was being deployed was still dual-core, and we happen to have the best dual-core solution.
RAM: And today?
JF: The world now is starting to make the quad-core shift, and by the end of this year you’re going to see a lot more of it. I was looking at the IDC numbers and was surprised to see that you saw a quick jump up for our competitor in the first quarter that they had quad-core and then the rate of adoption started to taper off. We believe that there’s a big market for quad-core, but there’s a much smaller market for a less than optimal quad-core. We think they’ve captured most of the less-than-optimal market and that the rest of the market is waiting for a native quad-core design.
RAM: I’m glad to see that AMD has backed away from bashing Intel’s front side bus design. The modern Xeon chips obviously haven’t hit a bandwidth ceiling, and I don’t know that AMD’s criticisms on this point were very persuasive after a while.
JF: I wouldn’t say that we’re not talking about the detriments of the front side bus. When you buy a car, you don’t really think about what kind of engine is in it. You look at the fuel economy, how fast it can go, and a host of other factors. Customers don’t buy a front side bus or Direct Connect Architecture. They buy an architecture that’s going to scale. So we spend a lot of time talking with customers about the attributes of their applications and what kind of scalability they need. It’s better to have that kind of discussion than say, ‘Whoa, that one’s got a front side bus. It’s never gonna make it.’ If you look at process technology, you’ll hear my competitor beat its chest about the fact that they’ve got 45 nm technology. But again, I’ve never seen a customer buy a nanometer. They buy processors that consume a certain amount of power and deliver a certain amount of performance. If my 65 nm product can beat their 45 nm product in performance-per-watt, the number of nanometers is irrelevant.
RAM: All right. If you know that Barcelona has Intel beat in given situations, where’s all the messaging? Where are the ISV endorsements and guidance about which apps work best under AMD?
JF: Benchmarks are such an interesting thing. Customers look to benchmarks as a proxy for what their actual performance is going to be, and every single system vendor is going to point to the benchmarks. I can always show you one that’s going to make me look good and my competitor look bad. But if you take a step back and ask what a platform looks like that’s going to deliver better performance in certain benchmarks, my competitor does well in SPECint benchmarks. The test essentially sits in cache and runs at clock speed. They’re always going to have a clock speed advantage on me, so SPECint is gonna run better for them. Now if you go out and ask somebody in a data center how many of their systems are running work loads that are very clock and cache sensitive, that don’t have a lot of I/O or memory bandwidth requirements, you pretty much are gonna get a blank stare. We tend to do well in scenarios that are heavy in I/O and memory, and that’s about 90% of what goes on in data centers. One of the areas you’re going to see AMD do a lot better in is virtualization. That’s a very hot topic now. Native quad-core gives us very good performance in databases. The larger the application, the more users and demand you put on it, the better off you’re going to be on Opteron.
RAM: So AMD says one thing and Intel says another. How can a reseller know which camp to trust?
JF: For us, one of the most important things is to provide our partners with sample processors and platforms that allow customers to really make the comparison on their own. In the big scheme, I generally win if I can run the same application on my competitor’s platform and mine, with my customers’ application and data. As long as they’ve got a power meter at the wall, that’s also the best way to see how it’s going to perform from a power-efficiency standpoint. In side-by-side comparisons, I have an extremely high hit rate with customers. For customers who only look at specs and benchmarks, we have to do more work with education.
RAM: Many of the architectural benefits that AMD has been staking as the key advantages for Barcelona over Xeon are about to disappear when Nehalem arrives later this year. Will this put AMD a generation behind?
JF: I’m not really too concerned about AMD being “a generation behind” when Nehalem comes to market because, frankly, Nehalem is Opteron 1.0. Welcome to the game. You finally integrated your memory controller. You did the native quad-core. A glueless interconnect. We’ve done all that and have been for quite some time. We’re not concerned about the Nehalem hype our competitor is throwing out there. We know that doing native quad-core is not easy. We know that integrating a memory controller is not easy. They did it twice with Timna and Whitefield, and they had to kill both of those projects before they ever got to market. The proof is in the pudding.
Now, where you’re going to see the true advantage of a glueless architecture with integrated memory controllers is in the four-way space. Nehalem in a four-way is going to be about a year late, like around the end of next year. Four-way may not be a huge part of today’s market, but with as many people as there are talking about virtualization, four-way plays very well there. In fact, Shanghai, our next generation of Barcelona, will be coming out around the end of this year. And Shanghai fits in the same infrastructure as Barcelona, so the channel will be able to take advantage of it immediately. On the Intel side, you’ll have to change your chipset, interconnect, processor cores. With Shanghai, you just have to qualify the processor and you’re off to market.
AMD ON ABC
RAM: AMD has several validation programs...and apparently two different AVS programs, yes?
RM: Yes. AMD Validated Solutions defines a validation suite, defines the service and support piece, defines specifications that are purpose-built, for example, for business platforms. AMD Validated Server is the barebones server program that is much more basic in foundation. It makes sure we’ve got barebones servers that have been validated and tested. We seek to address problems some people might have, like thermal issues and electromechanical things wrapped around the CPU. Validated Server is much more technical and focused on the “hard” hardware piece of it. AMD Validated Solutions is more the back-end process that really is addressing everything from the service and support to the specifications so we can develop purpose-built platforms that resellers can then put their secret sauce on top of.
RAM: Now you’re rolling out Business Class, which is essentially AVS 3.0?
RM: Right. Business Class takes all the steps and processes we’ve learned from AVS over the last three stability periods, and now we’re starting to apply that to very specific, targeted platforms. With Business Class, you’re going to start to see some very mainstream, business-related features. For example, the dual video out. We had dual video on the last go ‘round; we’ve definitely improved on it now. We also have TPM [Trusted Platform Module] headers so system builders have the option of using a TPM module. Business Class really equals the evolution of the program really based off of system builder feedback.
RAM: I’m a huge multi-monitor fan. While I might argue about whether people need a cutting-edge IGP to power two screens in a business environment, the fact remains that multi-monitor is a terrific value-add for channel resellers. And yet I don’t see any vendors stepping forward to help resellers promote and sell through on this functionality. Why not AMD? It sure seems like an easy target for you guys right now.
RM: I was talking with a couple of system builders just today about the ways in which multi-monitor gets used. It’s something we need to communicate more and do more education on. Business Class is the perfect opportunity, because our platforms have outstanding performance, whereas the other guy is challenged. And because of the advantages we have, that is one of the places we’re going to continue to press our competitiveness, and a key thing is enabling our resellers to use that as a selling point.
RAM: Setting aside board features for a minute, what else is different about Business Class versus AVS?
RM: We’ve extended the stability period to 18 months for desktop. We have increasing stability. There’s a much more commercial-centric specification. The other piece is we’re adding AMD Business Class-specific processors, which carry up to 24 months of longevity and stability. That’s another thing system builders requested.
RAM: AMD lobbed out a couple of headlines about DASH almost a year ago, and I thought for sure we’d see DASH in your new stability platform . . . but it’s not here.
RM: We’re really focusing on mainstream features. We talked with a lot of system builders about manageability, and we found that the majority of them are utilizing in-band manageability and not so much out-of-band. While some of them use it as a check box, most are not implementing it. So we’ve taken a lot of time to develop a manageability standard that’s going to be open. System builders in particular don’t want to get locked into a proprietary standard where, when switching, a customer has to start from scratch all over again. You’re not going to see a DASH implementation on this round of boards, but most definitely on the next round you will see some initial implementations. Again, this is based on what resellers are telling us they need.
RAM: A skeptic might say AMD is just waiting for the market to come to it on manageability rather than leading the market to a better solution.
RM: Not necessarily. We agree that remote management is something that definitely has great interest and impact for system builders, especially as the line between MSPs and solution providers becomes narrower. Solution providers are beginning to use that as a differentiator. Certainly, there’s an opportunity there, but we also think that system providers need to step into this carefully. If you’re killing a gnat, using a bazooka probably isn’t the best approach. When you have end-to-end control of a solution and ecosystem, you become myopic. Even if you try to expand, if you built a locked-down system from the outset, you’re stifling innovation.
RAM: AVS had a server component. Does Business Class follow suit?
RM: Business Class is very much a purpose-built platform. AVS had a pretty wide range, but what you’re starting to see is us getting specific on the features that target SMB mainstream customers. And yes, you’ll also see this on the server side. We’ll be rolling out more 1P boards in the near future, and you’ll see some specific features there that all make sense, especially for customers stepping into AMD for the first time.
RAM: If you had to encapsulate the last six months of news inherent to AMD, good and bad, and deliver a single statement to take forward to resellers, what would it be?
RM: When you look at the TLB errata issue, we took a very conservative stance on that, and a lot of that was based on doing the right thing for the channel. It’s about a balanced approach, not just in the validation of the processors but overall processes and making sure they’re working together to ensure that as new products come out, we’ve got not only the base architecture but the full infrastructure in place. System builders need that so they can execute right out of the gate.
| Back to top
Copyright © 2008 RAM Magazine. All rights reserved.
Do not duplicate or redistribute in any form.