Sunnyvale (CA) – If you have been waiting for some news on how AMD will counter Intel in the next one, two years, then the Technology Analyst Day provided many answers. Not a lot of finished products, but AMD outlined its ideas which the company hopes will shape the industry down the road. If you have missed the webcast, here’s our comprehensive wrap-up.
AMD held its semi-annual Technology Analyst Day earlier today. The primary purpose of this meeting is to convey technology information to the public based on AMD’s direction. Because of this the focus and tone of the meeting was only positive. Another similar meeting is scheduled for December where, in addition to the technology, a deeper financial outlook will also be center stage. (Read also: AMD gets more aggressive, announces next-gen CPU cores )
AMD provided us with information on future products, some revised products and even shared some of their insight into this currently evolving industry. Overall, the presentation was very informative, upbeat and provided a feel for the technology mindset of AMD.
Larger product base
One of the big highlights from the event was the large number of consumer products AMD will begin targeting. The firm’s slides showed 14 different classes of devices. To accommodate this wide berth, AMD has introduced two new technology platforms, called Bulldozer and Bobcat (like the skid loader, not the animal).
Bulldozer is a higher-end product aiming at server, desktop and notebook platforms, while Bobcat targets the lower end, such as ultra mobile computing devices. AMD sees these two technologies moving forward as the centerpieces for a whole fleet of targeted consumer electronics devices. AMD also showed us some Imageon and Xilleon system-on-a-chip products. Imageon already has shipped 2 million units. These products will also include new, specialized equipment for more features and better power savings.
The manufacturing goals at AMD have changed as well. No longer do we see independent efforts within AMD’s doors. Instead, we’re seeing a shift away from the more traditional manufacturing models aimed at particular segments. AMD is now looking to leverage its base compute abilities across the board. This means knowledge gained in one division is shared and incorporated in another. This includes not only CPU and GPU areas, but also more specialized compute abilities. AMD introduced a new UVD module (Universal Video Decoder, which works with Blu-ray or HD DVD), for example. The company stated that UVD allows a 25x greater utilization of silicon real-estate while simultaneously providing a 40% reduction in power consumption and a 40% increase in performance. These specialized modules are a win-win for consumers.
AMD’s focus has also evolved, and without question. They’re now taking design and manufacturing processes and leveraging them internally toward goals wrapped in “M-SPACE”. M-SPACE stands for Modularity, Scalability, Portability, Accessibility, Compatibility and Efficiency. It now permeates all aspects of design and is present in all forward-looking products (those about 2+ years out). It’s also the mindset in departments and technology ideas in general.
The modular design approach AMD is pursuing was mentioned consistently throughout the presentation. It allows a relatively small base of compute abilities, be they CPU, GPU, or individual components like memory controllers, HyperTransport, DirectConnect, etc., to target many more industry segments than would be possible without it. AMD indicated that by making their products more dynamic and configurable, they can expand them up and down relative to industry targets. This places its compute products in market segments beyond the base abilities the product might otherwise have provided for.
One of the big advantages AMD said it saw in this type of design was the common set of development tools. Also, the advantages offered by a type of backward compatibility from the base x86 architecture. AMD made a point of also not limiting its own future enhancements by sticking with x86 hardware alone. The company will be recommending and employing a layered software model across the board. This design is forward-thinking and allows the hardware underneath to change while still providing the same interface for existing applications. This will also be true in GPU products.
AMD explained that it will see x86-based product migration into nearly every segment, all the way from HPC to handhelds. The firm won’t do this by introducing dozens of specialized products, however. It will leverage their two base products, Bulldozer and Bobcat, and use component modularity to address the variables within those market goals.
AMD is looking at the direction industries are taking. The customers are speaking in the marketplace and AMD’s goals are to be there.
As such, the company is expanding existing products and looking for ways to leverage a smaller, yet capable base of compute abilities for new markets. While it might outwardly seem that AMD is moving away from its previous bread-and-butter technique of doing one or two things really well. Chief executive officer Hector Ruiz personally responded to one of the Q&A questions at the end which addressed this very concern by saying, “we believe that it is the product’s design which allows us to target this many markets.” He was referring to the modularity described above and how it can help increase the range of performance for varied processing needs across unrelated products.
Read on the next page: Manufacturing targets, Progress with ATI
As AMD explained its dual x86 vision, we saw two general categories forming. AMD called them the “factor of 10” approach. This approach indicates that a given design architecture can generally scale in power and performance by about a factor of 10 without significant redesign. This means at the low-end there will be x86 products in the 1-10 watt range (Bobcat). These will service products ranging from desktop appliances, high-end cellphones, digital TVs to other mobile products and more. At the higher-end, there will be the 10-100 watt range products (Bulldozer). These are for more traditional products, like desktops, notebooks, gaming machines, home media servers and even high-end servers.
AMD’s modular approach has allowed the company to take a step back from its previous models and design goals. The firm has analyzed and realized that there’s far more market potential for new products based on their baser resources applied through fundamental building blocks than was previously thought.
This was a powerful segment of the presentation and probably not only revealed to us what AMD sees as new, real potential. AMD’s design team now sees base compute abilities and internal technologies as fundamentally mature components – those which can be wielded to leverage new products more quickly. AMD was able to ramp its 65nm production capacity from startup to mature yields in about 1,000 wafer starts, for example, we learned. That speedy ramp was absolutely unheard of in AMD’s past and speaks very highly of the manufacturing process AMD has in place.
AMD showed a slide demonstrating that it took many 10s of times more wafers at previous process technology nodes to reach maturity. But now, thanks to much learning and the many manufacturing technologies AMD has invested in over several quarters with Fab36, it’s beginning to pay off. AMD can now wield design components at a much higher level than was possible before. And it is this new manufacturing ability, coupled to the realization of heterogeneous computing (through the acquisition of ATI) which has now allowed AMD to realize a fuller potential.
AMD spent a great deal of the presentation going over this new realization. During the Q&A session at the end, one analyst asked Hector Ruiz about the change: “Last year it was all about marketshare, marketshare, marketshare. What’s changed this year to place your focus away from marketshare and now on so many new products?” Ruiz replied that AMD has seen these new abilities emerging since the acquisition of ATI. The new compute technology bases being formed serve as the foundation of new abilities.
AMD also showed us how these new compute abilities are permeating into other industries. Hector and a representative from Microsoft both talked about how much the industry is in a state of change right now. One that’s absolutely unprecedented. It was very clear that the absolute hard-and-fast reliable goals and visions of just a few years ago are now being migrated. Those previous components might now only be pieces of a larger vision. And where we are right now represents only a mere fraction of the potential. Again, it all stems back to the very dynamic and flexible modular design process AMD is operating under.
Acquisition of ATI
One of AMD’s most strategic stated goals has been the acquisition of ATI. Ruiz indicated that AMD is completely happy with the ATI acquisition and transition. ATI brought a technology base to AMD that has served as a catalyst for new ideas and technologies.
Apparently, ATI’s intellectual property influx has allowed what was previously a possible vision to now become an attainable reality. The various representatives from different departments each came up and explained the potentials there, both for their departments and AMD as a whole.
AMD (with ATI) presented a unity and focus on the future. There were some what we would consider powerful product possibilities being discussed. The teams working on them claimed to be focused and committed very solidly to the end goals.
AMD also indicated how much performance there is to be had in the GPU. They indicated the GPU shader units will continue to scale, that these resources will be leveraged for additional compute abilities and that there will always be discrete graphics components as well as Fusion-like technology. When asked about this, AMD responded by saying that the GPU itself can do things in a specialized way that cannot be done in a more generalized way via a Fusion-like engine. As such, the need for high-end GPU compute abilities will remain a goal for AMD.
In one demonstration, ATI showed 1.5 teraflops of runtime computing ability for a live demo with a lot of demanding 3D effects. It was part of a game which will be coming out before too long and showed a stunning 3D flight through a forest, over some water, past a bird in flight, through the grass and much more. Another demo showed a single machine doing gaming and realtime MPEG encoding on the frame-by-frame generated images for the game. That information was transmitted across a network to be shown on two other computers. It worked so smoothly on the single machine that it almost could not be understood what was taking place. But, according to AMD, it was doing 100% realtime encoding of the images while generating them. The presenter made a point of indicating just how compute intensive this was.
Read on the next page: New graphics abilities, customer focus, commercial commitment
New graphics abilities
AMD hopes to provide “King of the hill technology in processors, graphics, chipsets.” It is working on a new video standard: DisplayPort. It will allow for higher resolutions, faster refresh rates, greater color depth and much more. AMD would like to see laser print-quality displays with crisper text and colors. AMD said it would also like to see the bulk of rendering bitmaps moved entirely off the CPU and placed in the GPU.
Additionally, AMD is working on new software technologies and techniques which shift the focus away from rendering images, which look real color-wise, to allowing software developers to create characters, which look real action-wise. AMD is doing this by moving away from the traditional “bone” model, where characters have a simulated bone structure which is then covered with polygons as they move. The new model was called a “muscle” model, and allows the surface to be rendered with many more programmable components. AMD is looking at ways to create a type of DirectX layer that would allow that kind of modeling to become the new standard as well.
The company is also looking at ways to enhance the frame post-processing to add effects which make it look more real, like motion blur. AMD is looking at ways to allow much more data to be processed to create higher resolution images with higher frame rates, more detail and more end-user realism.
The high parallel compute abilities seen in GPUs have made new technology ideas possible, like gesture processing. We have seen demonstrations of gesture-based recognition through physics simulations, for example. As one would move their hand through a virtual array of reactionary particles, they would respond appropriately. It’s the high compute abilities of the GPU that allow for that new input technology. It was indicated that the GPU can provide 2.5 billion instructions per second per watt. That is orders of magnitude greater than a general purpose CPU.
For notebooks, AMD is looking at PowerXPress. It’s a technology utilizing a dual graphics approach. For performance, a discrete GPU will be used. For long battery life the system will automatically switch over to a much less performing integrated GPU, but one which will work generally and provide longer battery life.
One final graphics ability worth mentioning is the direction handhelds are taking. AMD sees low-end notebook resolutions (1024×768 or 1280×768) soon arriving in handhelds and high-end cell phones. These new devices will follow the current trend of flash and pizazz. As such, AMD will be developing graphics products to meet their low-power, high performance needs.
Another of AMD’s consistently stated goals is to be customer driven. AMD reiterated that it is always listening to its customers and make strategic decisions to address their needs. Two direct customer requests have been integrated into their new, future goals. The first was a repeated request for AMD to design, develop and market its own server-based chipsets. AMD confirmed that is now being done. The second relates to maintaining the collaborative effort approach AMD has already embraced for many years. This is through partnerships and consortiums. AMD also reiterated its commitment to open platform architectures and solutions.
Customers apparently have also told AMD it needed to have more IP. It looks like that this was one of the driving forces behind the acquisition of ATI. It started with the idea to have a full arsenal of compute abilities at their disposal. The purpose of that goal was to move forward with a wide technology base capable of impacting compute abilities across the board. These would address the customer’s wants of having a “single stop shop” for the whole gambit of machine hardware resources, AMD said.
AMD took some time to tell how it wasn’t very long ago that its corporate client base was quite small. But today, nearly every major manufacturer produces AMD-based solutions.
The commercial world now sees a large enough base that AMD’s presence is definitely felt. Ruiz indicated AMD products are now being proudly marketed. He contrasted this with the past where AMD’s product name would almost be “dangled there at the end,” almost apologetically. Today, AMD is recognized as a solid commercial base and Ruiz made the point of stating how AMD is very proud of that fact.
AMD’s commitment to the commercial customer was also shown by its recent G3MX memory doubling architecture. We learned today that this technology not only promises to improve memory capacity, but performance as well. Going even further, AMD is looking at introducing faster compute engines inside their chips to accelerate IPCs (Instructions Per Clock).
AMD said it will be adding new hardware abilities for virtualization, I/O virtualization, extended instruction sets and tools to its products, which will help the multi-threaded developers monitor performance inside the chip during runtime execution. These new sets of tools will allow a developer to get a real hands-on feel for how much utilization is seen, what can be tweaked. AMD will also provide insight into how to tweak it.
Read on the next page: OLPC, Process technology, Author’s thoughts
50×15 and OLPC
AMD renewed its commitment to the 50×15 initiative, which would like to see 50% of the world’s population connected to the Internet by the year 2015 (only 7.5 years away). Ruiz made a point of indicating that AMD’s initiative in the OLPC (One Laptop Per Child) is now gaining momentum. This was a veiled reference to Intel’s recent adoption of that program as well.
AMD indicated it is on track with 65 nm. Yields are higher than they expected. So high, in fact, that they were able to push back some capital expenditures. AMD’s Fab36 facility is at full 65 nm production right now. AMD is also running pilot 45 nm products through. Fab30 is scheduled to be retooled into a 300 mm, 65 nm facility an will be called Fab38.
Right now, AMD’s slides indicated that it will be in 45 nm production in the second half of 2008. There was some confusion about the availability of 32 nm processors – while some slides indicated that volume production of 32 nm CPUs would happen in 2010, some 32 nm processors were listed in the 2009 time frame – which would be a tremendous achievement for the company. 22 nm production was shown in the 2011 time frame. AMD also indicated that high-k/metal gates may appear in the second generation of 45 nm processor, but will definitely be present at 32 nm.
AMD will be migrating its graphics technology from 65 nm to 55 nm in 2007. It will be using an Asset Light model, which was described as a way to maximize all available personnel and equipment, involving third party bulk production facilities at Chartered, UMC and TSMC.
Hector Ruiz came out at the end of the analyst day and gave a summary of the presentations given throughout the day. At one point in the summation, he noted that AMD’s management is “not confused or distracted.” Everything AMD is doing is in line with stated business goals and fundamental concepts which, over time, are coming to fruition.
I’ve been a little concerned about AMD in recent months. This has been especially true with the firm’s huge quarterly losses. They have low cash on hand and a product line which, from all outward appearances and early reports, doesn’t seem able to compete with Core 2, let alone Penryn. But I must also admit that a lot of that changed when I listened to AMD speak about what they’re pursuing.
I’ve always thought of AMD as a desktop/server/mobile chip making company, pretty much in that order. But after listening to the presentation today, it appears to me that AMD has refocused its lens. The compute models alone, for example, are geared toward what they’ve been claiming all along: Performance-per-watt advantages across the board. Those goals actually include the ability to create custom processing units which drive compute needs and best serve the customer.
AMD has done an overall good job to restore confidence. It removes much of the haze after three huge, consecutive quarterly financial losses. This presentation showed a company that was capable, on focus and extremely innovative.
The only question which still remains is whether or not the vision can come to fruition in the marketplace and if AMD will have the cash to do it. Your guess is as good as ours if that will be the case.