Sunnyvale (CA) – The opportunity is huge: A personal computing platform with low power consumption, designed for mobility, merging a high-performance 64-bit computing platform with the parallel processing power of one of the most capable graphics platforms ever developed. When you see the prospects in terms of the fully assembled jigsaw puzzle that they jointly announced on Monday, it may appear that AMD and ATI have managed to fit all their pieces together. At last, a newly reinvigorated company will have a complete, competitive response to Intel’s Centrino.
It’s when you ask both sides of the merger agreement to assess the inventory of all the pieces they plan to assemble, as TG Daily did this week, that you start to notice some subtle, but pronounced, differences in the two partners’ view of this colossal opportunity. It’s as though we were to follow the story of the two characters in the TV commercial after they had accidentally discovered that their peanut butter and their chocolate tasted great together: Their new objectives have the same basic premise, but there’s a disparity over which part gets to be the creamy coating and which the yummy nougat.
It’s a question of which part goes on the outside and which part is subsumed, in a sense, in the middle. This week, we spoke with high-ranking representatives from both AMD and ATI, asking them to describe the new platform they have in mind. There don’t appear to be pieces missing; instead, the problem appears to be some pieces left without a place in the puzzle after it all comes together.
The whole system perspective
From ATI’s perspective, according to representatives we spoke with there, that company faced a serious evolutionary hurdle: In order for graphics processing to evolve, the underlying system had to evolve as well. And if ATI didn’t have a voice in this process, its product line would have nowhere to go but sideways.
“It’s rather like making a car by just focusing on the engine, right?” posed Chris Evenden, ATI’s director of public relations. “There comes a point in which making the engine better doesn’t actually make the car any better…At that point, it starts making more sense to look at it from a whole system perspective.”
“There comes a point in which making the engine better doesn’t actually make the car any better.”
Chris Evenden, director of PR, ATI
Intel’s Merom architecture – which extends its Centrino mobility platform – lowers the power consumption on notebook systems considerably. AMD needed a response to that; and as Evenden explained, some of what had been considered AMD’s architectural virtues may have proven to be obstacles to that end. With memory having a direct connection to the CPU via the HyperTransport bus, he explained, every time an ATI graphics processor placed a directive to refresh the screen, memory would need to be accessed, and that means going through the CPU.
“That’s an issue, because keeping the CPU awake requires more battery power,” said Evenden. “You could conceive, in the future, of integrating the GPU into the CPU so when the computer’s sitting there in idle – which is pretty much most of the time – you can switch off the CPU all that time where it isn’t being used.”
Both companies speak of “integration,” not on just the obvious financial level, but the architectural level as well. ATI foresees a day where it can be part of an organization that can change any part of the PC platform it wants to, in order to develop an innovative product. ATI’s Senior Vice President of Marketing, Rick Bergman, refers to this as “the platform strategy” – the joint vision that his company and its future partner see themselves driving toward.
“As we look out in time,” Bergman told TG Daily, “[we see] the platform strategy becoming more and more dominant. The PC customers that we have, the end consumers, are looking for solutions at that level, partly because the problems that we’re encountering now are tough problems, like the power issue. You can try to solve it the way AMD does today, where they get different pieces from different companies, but that certainly doesn’t create the best solution, nor is it totally efficient as well. So once we can control all these pieces and come up with the best solution, it really changes the dynamic. That’s how we saw it, and that’s the same vision that AMD saw as well. We are totally aligned there.”
Bergman’s colleague, Chris Evenden, put it this way: “If you think about the PC architecture, it’s been fairly stagnate for several years. There’s been little significant change to PC architecture for awhile. The last big change that people point to is PCI-Express. You look at that and think, ‘That was just a new bus protocol.’ It’s hardly earth-shattering. It didn’t conceptually change the architecture in any way. Whereas, now having [CPUs, GPUs, and chipsets] under one roof, we can actually change the architecture, and still make it a PC. It’s still just as programmable as any other PC, but we can really customize it for certain applications.”
The rise of the ODM
The messages from ATI and AMD are alike in this key respect: Both told us their customers are demanding a single “point of contact.” As OEMs and ODMs – with the “D” now standing for design – strive to build notebook computers at lower costs rather than desktop computers whose buildouts change weekly, they want to be able to know exactly what to build to meet a specification that end customers will accept.
“Once we can control all these pieces and come up with the best solution, it really changes the dynamic.”
Rick Bergman, Senior Vice President, ATI
AMD marketing architect Hal Speed characterized his company’s take on this customer demand: “One of the things that the OEMs kept telling us was that they wanted more of a complete solution from AMD: chipsets, platforms, reference designs. Not to lock them in [to one spec] and use business practices that force them to use that, but to have one point of contact, to have one reference design they could start from, and then differentiate in their own right, to not have to negotiate separately with each chip vendor on longevity and image stability, and things of that nature.”
“There’s a quick win there in terms of market share on notebook and commercial desktops,” said ATI’s Chris Evenden. “AMD is woefully under-represented in those markets, and largely because they can’t offer a complete platform that those customers really want…The customers are [saying] they want to see AMD with a chipset as well. That’s just a quick-term win, and that’ll allow the bit that used to be ATI, to sell more chipsets.”
Obviously, the customers to whom Evenden and others refer aren’t Alienware or Falcon Northwest, but the burgeoning field of ODMs, even the leading members of which have names few have heard of – Quanta, Compal, Wistron. Collectively, these outsourcing firms are responsible for 82.6% of all mobile PC production worldwide, according to iSuppli estimates released Wednesday. These are companies that are collectively driving CPU manufacturers to develop standardized reference platforms that enable them to engineer production processes that maximize workflow while cutting costs.
Intel has enjoyed the advantage of being both a CPU and chipset manufacturer, with some prowess with integrated graphics. It hasn’t produced high-performance graphics processors; but as it’s turned out, Intel’s lack of recognition in that space has not been to its detriment. For AMD to compete, it needed integrated graphics, and it needed it yesterday. So spending too much time doing further research and innovation was out of the question.
“Let’s assume no architectural changes to the platform: AMD’s integrated northbridge, networking components, HyperTransport bus,” proposed AMD’s Hal Speed. “[OEMs] are looking for AMD to deliver more of a complete reference design for today’s commercial clients and notebooks.”
For this reason, ATI’s Bergman explained – and perhaps for this reason alone – AMD approached ATI. In other words, ATI didn’t suddenly decide it needed to expand its presence in the chipset business, and start shopping itself out to CPU manufacturers.
First, let’s get rid of all the lawyers
At the same time AMD approached ATI, Bergman said, “The executive team of ATI went through a strategic process, and looked at the future of computing, and how can we maximize the growth and innovation within our company. Certainly, we could have continued on our current path, and grown our consumer business aggressively. But actually, when AMD approached us, we said, ‘Wow, there’s a lot of things that make sense.'”
But why wouldn’t something on the order of an exclusive strategic alliance have made even more sense? Couldn’t two companies have agreed on participating in developing a joint platform, without all the burden of splicing their operating units together?
“The customers really wanted one company, and one point of contact, to be responsible for the delivery of the solution.”
Hal Speed, marketing architect, AMD
“We actually considered that, and considered other opportunities as well,” responded AMD’s Speed. “We could have developed our own chipset capabilities internally, or we could have formed some kind of partnership or alliance or licensing arrangements for a third party graphics chipset. But the customers really wanted one company, and one point of contact, to be responsible for the delivery of the solution. Then longer-term, ultimately we believe that there is going to be some silicon integration – not right away, but in the future – where we would have had to own the IP to really make that successful.”
Ah, to cite something one of Shakespeare’s endangered lawyers may have said, there’s the rub. It isn’t so much that AMD wanted to own ATI’s intellectual property. It’s that the licensing of IP would have been a pain in the neck. “There are things that AMD could be innovating if we had access to the GPU IP,” Speed admitted, “whether that’s innovating in a discrete fashion or innovating in an integrated fashion – meaning, putting it on the same piece of silicon. In our long-term plan, we knew we needed to have access to the GPU IP.
“Let’s say there’s something AMD wanted to do, and needed the third-party licensing company to implement for us,” Speed said hypothetically. “We would have had to go back and negotiate, ‘How much is that going to cost?’ and [say], ‘Gosh, that doesn’t really fit with our development plans; we’re really going in this direction strategically, and the direction you’re wanting us to go is not in line, so we’ll get to it when we get to it…’ This [merger] allows us to integrate the visions by integrating the companies, and then align the development resources around the common goals.”
It could therefore literally be seen as a cost-cutting move for two companies, rather than go through all the bother and expense of licensing each other’s technologies to one another, to just merge and get it over with. Furthermore, as Speed pointed out here, when two companies form a partnership or joint venture, their diverging interests often get in the way. It’s sometimes better to simply sew those interests together, one way or the other.
ATI’s Rick Bergman agreed: “The challenge is, at the end of the day with joint ventures or strategic engagements, they’re supposed to be servicing their shareholders and we’re supposed to be maximizing value for our shareholders, so you get strategically mis-aligned,” he said. “The depth and breadth of cooperation that we wanted, I don’t think you can handle with a JV [joint venture] type of structure. We’re talking about different levels of integration, we’re talking about forgetting the artificial boundary that exists there on how you architect parts, removing that and getting the thinking and implementation that we wanted to achieve the performance, power levels, and features and functionality, I don’t think you can do without coming together like we are.
“Really, it’s an opportunity for us to help participate in defining the PC platform, and to really be part of a great semiconductor company,” continued Bergman, “and not having other people come up with the rules for the platform.”
Can the technologies will merge along with their companies?
What role does the GPU play in an emerging AMD platform? Now that AMD will be a producer of graphics chips, what will they do, and who will they do it for?
It isn’t as though AMD had this burning desire to be building Radeon graphics cards – in fact, as we learned, that possibility may have been the furthest thing from AMD’s mind, and might not have even entered the discussion. Foremost on AMD’s mind was the need for an integrated platform, complete with CPU, chipset, and graphics. AMD’s Hal Speed even warned us that this news might not come as exciting to the typical reader of Tom’s Hardware Guide. But there are long-term concerns as well, particularly with regard to the emerging role of the GPU as more than just a graphics processor.
“To a certain degree, we won’t actually care where the processing occurs, whether it’s the GPU or the CPU, and that’s really the strength.”
Rick Bergman, senior vice president, ATI
With a platform where the CPU and GPU work more cooperatively, conceivably, the CPU may be better able to take reiterative or scalable operations and pass them off to the GPU, even if they aren’t graphics-related. “There’s the whole concept of what’s been referred to as GPGPU, what I think of generically as putting more tasks onto the GPU than just graphics,” related AMD’s Speed.
As ATI’s Evenden said, GPUs today are more flexible than ever, fully programmable, and will soon meet the IEEE’s basic requirements for general-purpose, 32-bit processing. “We’ve got both types of processors under one roof now,” he stated, “and each of them will find their own applications. There’ll be general-purpose computing that runs best on a multithreaded CPU, and there’ll be other general-purpose computing that runs better on a massively parallel GPU-like architecture.” In fact, ATI is preparing a demonstration of this discovery for just a few months from now.
Physics calculation is one example, mentioned Speed, of reiterative tasks that could conceivably be shuttled to the GPU, where it’s computed in a parallel fashion. “It gives us the flexibility now to not really be constrained by the limitations of each of the architectures,” he said, “and look more holistically across the entire platform and say, ‘What is the best way to optimize this hardware for the applications that people want to run?'”
As ATI’s Rick Bergman perceives it, co-opting the GPU for general purpose tasks is one of the new company’s first priorities. “There’s some areas that we both have strengths in,” he told us, “[such as] GPGPU and physics. One of the challenges there, is having platform access to some of these things. Immediately, with a joint company, we address the platform question.”
Here is where the two companies’ messages, prior to their being sewn together, start to diverge. Bergman believes GPGPU could drive his platform strategy in two directions, perhaps simultaneously. “You could see like an ‘ultra notebook’ type of platform, where you could just integrate the functionality together, graphics with the CPU. There, you totally focus on power. Clearly, integration has its benefits there. And as you move up the food chain…with these high-power GPUs, can you do new tasks, or tasks that have historically been done on the central processor unit? The answer there is, yes. These are huge, parallel, massively computational engines that are capable of doing a lot of different types of algorithms that have historically been done by parallel CPUs.”
Is there a place for the GPU in tomorrow’s platforms?
Bergman also explained that the segmentation of the PC market is changing, and as a result, how we perceive it should change as well. Conceivably, reforming PC architecture around new and emerging platforms could enable vendors to segment their markets completely differently – not toward such common segments as “consumer desktop” and “business portable” today, but in a much more targeted fashion. This could be because neither the underlying architecture nor the overbearing form factor of the PC would dictate to vendors who their customers should be.
“To a certain degree, we won’t actually care where the processing occurs, whether it’s the GPU or the CPU,” Bergman projected, “and that’s really the strength. If you think about it, AMD will be the only company that’ll be able to have that scalability all the way down to simple integration, up to many CPUs, many GPUs, to control both the hardware and the silicon, as well as the drivers and application sets on top of it. From that perspective, it could be very powerful.”
Not caring what’s doing the processing, might not be part of AMD’s bigger picture. Hal Speed pointed out that the GPGPU development and the merging of CPU and GPU, are two separate development paths, the former of which presumes that both devices remain separate. Merging the two processors may make sense, he believes, but only to a limited extent, and for a limited customer base. “There are opportunities, certainly, in the growing countries like India, China, and Brazil, to have an integrated solution that’s not legacy technology, but is current technology, but integrated in a cost-efficient manner that better meets and serves the needs of some of those markets and geographies.”
So an “emerging markets” program, on the order of the $100 notebook PC that gets so much attention, might be the best outlet for a merged CPU/GPU from AMD’s perspective. And even then, the graphics power we’re talking about wouldn’t be very impressive. “It doesn’t make sense to me to take a high-end graphics card at 200 million transistors,” Speed remarked, “and try to wedge that in with a CPU and a bunch of cache for today’s mainstream and performance PC market.”
“It doesn’t make sense to me to take a high-end graphics card at 200 million transistors, and try to wedge that in with a CPU [on the same silicon] for today’s mainstream and performance PC market.”
Hal Speed, marketing architect, AMD
The GPGPU project, added Speed, requires a different mindset, a different group of engineers…and a separate timeframe. Rethinking how to address computing tasks may yield a solution that’s integrated into designs a few generations from now, he told us. He called this “the silicon effort.” The immediate task, however, is for AMD to utilize the tools and IP the combined companies now already have, to build a reference platform that puts AMD back on the map with its ODM customers.
Today, ATI sees the merger with AMD as critical to finally being able to wrest control of the entire PC, to be able to shape all of it to suit its architectural goals, making compromises with no one. It sees this move as potentially elevating the role of the GPU to a position alongside the CPU in both priority and stature, such that the chip responsible for processing any given task may become inconsequential to most users. For AMD, the merger is an acquisition of a key element of intellectual property that a platform producer would need. It enables the co-opting of the GPU for purposes neither company might have been able to undertake, even acting jointly. But that’s just a bonus, an extra payoff that may come after AMD integrates the GPU into its platform strategy.
Both are worthy goals. Both use the same terminology and the same tools. The question at hand is whether they may coexist. For this reason, the two companies may need to come to a closer agreement on long-term strategy before the ATI in AMD ends up like the AOL in Time Warner.