AMD reveals its future plans

IN THIS SERIES of articles, we will be addressing current, near-future and future outlook of AMD.

Since the chip maker recently swallowed ATI, the company is sending out a few of mixed signals. Faced with tons of questions about the present and future of AMD, from our readers to industry analysts, this five-page-series will disclose what is going on with The New AMD, Daamit.

We open with an extensive interview with two high-ranking AMD officials, who gave us an insight into the way AMD is heading right now. After that, a series of articles will analyse the product line-up, technologies and the internal organisation.

INQ "With the AMD merger, there has been a lot of turbulence, so could you to explain your roles in the New AMD to all of the readers on The INQUIRER."

Phil "My role is Chief Technology Officer for the now combined company. I work as head of the group that works on strategies that deal with what technical solutions we should be developing. That seeds into select process that we use to pick areas to focus on. Those areas are selected by the business units inside the company, while Bob will talk about how ATI is organised - but, you might expect some small adjustments.

"We've got traditional notebook, desktop and server areas, but now we are combined with ATI, we also have to supervise a whole new family of devices and need to address their markets, etc. Senior technical staff I got, work on platform architectures, what technical diligence the acquisition requires and work on technical mentality - which is how AMD makes a great place for technical ideas to work. In a nutshell, this is what I do.

"I work closely with CTOs like Bob and heads of business units. We got a decentralised model that calls for as much customer-centric and competency in individual business units, so we really try to develop skills within those groups. My office is a kind of umbrella that combines the activity between the individual units."

Bob "I am now a CTO of Graphics Product Group at AMD. At ATI, I was the CTO for the PC Business Group, but now with AMD taking over, the name has changed a bit. The way I work remains the same, however.

"The Graphics Product Group is fairly autonomous. It is our responsibility to determine the right roadmap and the graphics processor space across the whole range. The full product market segment shows that we are diversifying in discrete graphics processors, the ones using the PCIe slot, and integrated chipsets that we see paired with an AMD or another processor. Our business, the Graphics Product Group rarely changed focus while we were a part of ATI and this will remain in future, only by expanding the integrated area."

INQ "Could you talk us through the creation of DAAMIT, or The New AMD as it is referred to as of late?"

Phil "Two or three years ago, Hector (Ruiz, CEO of AMD) started to think of taking AMD to the next level. We have an internal to-do mantra that goes: "To go from a good to a great company". The question was: "What do we need to do to get to there and how will the industry evolve..?" One of the things that came up pretty clear with was that the x86 (instruction set) is going to affect a wide range of markets, from handheld devices - all the way to super computers.

"One thing was pretty clear: media and graphics processing is going to be as important feature just as traditional data processing is. As we took a look, we thought about trying to do it (CPU-GPU) internally and came to the conclusion that a bunch of companies have already tried to combine the two. Those were really good graphics companies and really good CPU companies, but the risk and cost of going there was high.

"From that, we concluded that the right thing to do was to actually figure out how to partner with someone who is really good in graphics and media area, that also had competency in the lower-end, kind of consumer electronics space. Long story short, that ended with us and ATI starting talking together, around the first of the year, and from that point on, we concluded that our point of view (where the industry was heading) was very similar and complementary to the one they (ATI) had.

"We spend the last six months figuring out the details of how we will work together, what the roadmaps were, the business side the companies worked on, details of the merger, and long story short, we recently signed a four billion Euro check to finalise the transaction.

"So, what we are going to talk about here is the vision about what comes when you (have) broken down the barriers between the CPU guys, the GPU guys and get those things together. And the effect that will have not only on the PC stage, but also Bob will tell you the examples where this technology will show up in the handheld area.

The expression we use is: "Building from palmtops to peta-flops class systems". So, this kind of sums up where we are going to end."

Bob "I think that I can offer you more information from the perspective of ATI. I suspect that the graphics side of the talk we are doing now is the side you're mostly aware of, but we want to give you the sense of direction we are heading, once again from the perspective of ATI. Even though ATI has a commanding share in the handheld space, not many people are aware of our position as a leader in the segment. I am not sure are you aware about our graphic efforts outside the desktop space."

INQ "When it comes to cell phones and PDAs, the problem is that you cannot promote the brand and it is not focused on actually seeing how many GPU features have been added from yesteryear."

Bob "...Right. We have been impolite regarding the way we handle the promotion of processors in the handheld space. That line will continue as a part of AMD, as this is a very high volume segment. We shipped over 100 million products and we plan to double the number in this year. So, it's quite a big market and quite an important one. You know the type of features that have been asked for in that space. Everything that comes from the PC, everything that comes from consoles, everything that your digital camera records… everything that needs to be done on a single chip."

INQ "Yeah, but when it comes to shipping numbers, it is unknown that ATI ships far more products than AMD actually does. AMD does some 45-50 million CPUs, and now with Spansion ending up as a separate company, number of flash chips is no longer part of the numbers sheet. At the same time, ATI ships graphic chips, Xilleon, Radeon, with the mobile part, Imageon - just Motorola and Samsung shipped more handhelds with ATI chips than AMD's makes CPUs per year."

Bob "But that's what makes these two companies complementary. As you've mentioned…we do make large shipments of chips, the number of product releases we have per year is quite large compared to what AMD had. And yet, there are qualities that AMD has that we need in order to be more competitive on the market, and to keep our partners in the loop with the new technology. And so, the combination of two companies is to be able to integrate work in high volumes in order to be able to customize chips for particular segments, and by that we mean taking advantage of high-speed circuitry, understanding how to add computation to a quality level that is expected as we do more and more general-purpose type of applications. With more compute on a GPU, we expect that people will expect to have the power of processors on the market today. So, to me it's quite a good mix. There is very little overlap in what we did, so we did not fight about the right way to do things, it's really that we both need each other's technology."

INQ "When it comes to handheld market per se, AMD was present there with flash memory, while the ATI is basically the key player in mobile segment. That segment is changing a lot, with advanced technologies coming up in 2007 and in 2008. How do you see now these two companies satisfying the needs of companies such as Nokia (ATI has the role as a key player in future N-Gage plans), what will happen with this? Is AMD just going to "insert" its technology and enhance that platform or...?"

Bob "First of all, the consumer group that developed the Imageon chip is alive and kicking. The Imageon will continue its role as the media processor of choice for advanced handheld devices. I think there is a lot of things to discuss about what kind of processors could do well in handheld space, but to the extent that the vendors currently have specific choices about how do they want to use their compute. We will not bring down their compute, their processing power nor will we force partners to take something they don't want. We will offer partners the opportunity to develop AMD-type processors for this space, but if they want immediate processing capabilities, today, they will want to continue to use other partners."

Phil "If you're looking at current situation, when you see the companies such as Texas Instruments or Hitachi, they are all successful. What that says to us today is that you can't really differentiate on the terms of architecture alone. That's the stuff that makes the difference, as we continue to offer more levels of integration. With 65nm and 45nm manufacturing, the cost of transistor will be cut down, so at price point, you're looking at more and more transistors. That says again, if you're looking at a certain price point, you get more and more MIPS (Million of Instruction per Second) of power and that means you can run more and more complex software. At some point, the software stack will grow to the level of something like UMPC. And when that time comes, you will not need something powerful that you would like to run in a specific budget.

"But, that is going to be driven by the customers, and the way they are doing in terms of their applications. We will not try to force anything on them and the advantage that we've got is that when they reach that tipping point, when it makes sense for general purpose architecture - we've got both elements now. The development will depend upon how fast they can start touching the UMPC. Yet again, the UMPC of today does not come with the right battery life, usability or capability. At some point, they will become feasible. And when they do implement the features that are now present at cell phones, with the x86 processors, a whole new world of usage models will open.

"But we are going to be driven by innovation, and when partners asks us what to do, at the right time there will be the right answer."

Both AMD and ATI are aware that their brand awareness in the handheld area is too low for a dominant market leader. This is also what worries AMD the most - how will players such as Motorola, Samsung react to all insecurities AMD and ATI have been voicing outside the corporate walls. Nokia's future N-Gage platform also depends on cooperation between Nokia, ATI and companies such as Texas Instruments. However, mobile 3D chips running with AMD's transistor tech? Top cell-phone vendors wanting handhelds with newly developed low-power low-cost AMD CPUs running x86 instruction set? It may be dream come true for handheld industry.

INQ "Right now, you are in neat situation: AMD, got a decent stack of IP that Intel does not have, while ATI guys, that had so many problems in chip design when it comes to power consumption...got something very interesting. Basically guys, what will happen once you took out the best graphics tech and combine it with CPU manufacturing tech? Do you see GPUs coming our way with the clock speeds similar to those AMD is currently running?"
Phil: "Oh yeah."

Bob: "By combining the proper technologies many things can be achieved."

INQ "Then you can make CPUs and GPUs altogether (in a same Fab?)?"

Bob: "I don't think that's impossible."

Phil: "Actually, there is a place for CPUs and there is a place for GPUs, together and integrated."

INQ "Is there a realistic way to expect the future that is be consisted out of let's say, taking an multi-core CPU and putting three or four GPU chips for HPC space?"

Phil "You must have seen our presentations (laugh)".

Bob "You are talking about multi-chips. What we are saying is that you take a multi-core processor, with its heterogeneous type of capabilities, and put it into very complementary environment."

INQ "Let's get back to the IP part. AMD now has the licence to produce Intel chipsets. Will AMD indeed continue to do so and what will happen with future chipsets for Intel platform? Are they cancelled or not...?"

Bob "Short-term, we are going to continue to offer ATI integrated chipsets for Intel processors. Long-term, we will have to assess our business strategy and talk to our partners and to Intel, to see is it financially feasible to continue to offer products for other processors. Technologically wise, there would not be any problems, only open question is the one of receiving future products (for development) from competing companies."

INQ "Microsoft is now coming out with Windows Vista and DirectX 10 API. I know for a fact that a lot of vendors are expecting a lot from the fusion of you two. How do you actually see AMD moving forward in 2007 and 2008? What will happen, what can you offer now that you were not able to offer before?"

Phil "To us, once you get to Vista, it requires a level of 3D graphics that Microsoft has never required. Analogy would be similar to those from mid-80s of the past century, once you saw 80386 and 80387. You know when you ran spread-sheets, the speed at which 80387 was doing that at the speed 386 could have only dreamed for. Same thing happens here, the 3D is now in its seventh generation without the application that would use it in a way application was using floating-point on the coprocessor. We think that by implementing a GPU into the die for 2008, the 3D will reach that tipping point that it really does make sense to implement the GPU inside the CPU silicon.

"Once you do that, we think we will have enough compute power, because it is behind the cache hierarchy and all that, notebook and certainly entry-level desktop will come with integrated graphics that will provide a good 3D experience. On the desktop side, this does not mean you're stopping to have a PCIe slot for discrete graphics. By then we could reach the point when economic equation seems that everyone who pays for mid-range and high-end hardware continues to do so, but certainly the entry level will offer sufficient level of performance for many consumers of the future.

"Now, in order to do that we need to start making GPU instructions more generally available - so if you take a look how did we took those 64-bit extensions and made them available for everyone, we managed to have all those binaries continuing to run at same or faster speed."

Bob "Other things that are important to appreciate are that we learned how by joining together, we can increase our efforts on all sides and those efforts continue. We did not stop with our development, we didn't wait to figure out what we will do now, and we're one company. We are executing for the DirectX 10, making sure they have the best processors, best chipsets, and best graphics. In the short term, you'll see more solutions that are focusing on integration, but largely we will continue to make the best graphics processors we can, and the microprocessor group will make the best CPUs they can.

"In the longer term, I think that you will see pre-set situation is changing and how the computing is changing, evolving - there will be more flexible solutions. But in the short run, we will gain the knowledge and expertise that AMD has for high-speed compute and we will introduce AMD to be able to do integration of parallelism and stream computing."

INQ "Favourite topic of the past couple of months: Torrenza. Do you see Torrenza being used as combining the CPU and the GPU platform in anyway together, like offering a high-performing GPU, just by using a socket?"

Phil "Torrenza means to me as being a complementary solution in our platform integration. There are advantages and disadvantages of integrating stuff on a die: the advantage is certainly it is often the most efficient way to do it, disadvantages - everybody has to pay for it. So, unless a broad set of applications ca use the feature we put on the die, it's not a good business equation. The reason is that you're adding the cost and people can't take advantage of. What we have seen is the server space, is a broader, a far more extensive range of different workloads, so transaction processing encouraged us in a way. But when you see new applications like Java, video encoding, you know, those sort of behaviour and those may be like 10% of total workload. So, what makes sense is to build another co-processor that is optimized for those applications to take the edge of the main processor. And so, Torrenza basically enables that so now companies can make specialized systems for markets, an incremental piece that plugs into the interface. And so, what Torrenza lets you do is addressing all of these 10% markets by creating a common interface but at the same time being able to have a coherent memory interface part, and have a full processor with the same capabilities that internal purpose processors have for the specialized applications. In many cases, it gives you excellent price/performance ratio and also significant improvement in power performance in already mentioned special applications.

"And so, we can ask for many things, we can make a GPU that plugs in Torrenza socket, from a technological standpoint, you can absolutely do that. But it's unlikely this is exactly how we would use it, because there is a system architecture around the PC in terms how the CPU and GPU talk together, over PCI Express and the performance you get out of the GPU is largely limited by the dollar of power budget. With that same dollar in power budget it is not clear whet ever you get better performance as a Torrenza co-processor. It would also require a brand new programming model."

AMD knows that right now, it's financially unfeasible to offer GPU on a socket. However, there are many challenges for GPUs of today to become a part of 1U and 2U rack servers. It all depends on will Google, Microsoft or Yahoo will pick Torrenza up for their massive video search engines. Let's dig into that issue...

INQ "AMD has been a straightforward company which led the GPGPU initiative in the movie industry (reference to AMD-Nvidia deal which brought Quadros in the world of movie FX business), by rendering Episode I on Opty-Quadro box on Opteron launch in 2003 and now being a standard in many FX studios. ATI recently launched Stream Computing initiative... right now, the GPGPU market is actually opening."

Bob "...right."

INQ "So in order to enable the GPGPU platform, some modifications in PCB design would be needed. There are some rumours that AMD could now start to lead the industry effort in order to accommodate the push of GPGPU and x86 instruction set in the HPC arena."

Phil "We'd love that. In fact, this is the very first step of putting the CPU and the GPU together. There is a huge parallel in our own visions of future and what guys in the companies like PeakStream and universities such as Stanford are doing today. Applications simply work with the GPU and this enables unleashing the GPGPU onto the world like oil field simulation and financial services. There is some really exciting potential, but we have to understand that there are some limitations in terms of GPU still being a fairly specialized unit and that software can only go so forward with the restrictions that exist on the GPUs of today. Once you take that GPU capability and put it on die, make the instructions more for the general purpose, and allow the GPU to do the workload of more general-purpose computation, then you can expand that PeakStream well beyond what you are able to do today.

"So, to us, GPGPU is first kind of step in enabling that capability, but again, you can only go so for with the specialized chips like GPUs of today. And so at some point, when you integrate onto the die, you can have literally, that performance on a single place."

Bob "In Torrenza, AMD is really being open to non-x86 standard processing, supporting the types of processors that are out there, like the vision Cray and IBM are building right now. They are compute processors that will cooperate with standard AMD CPU, so I am not sure why would AMD want to pull away from that. They are the company that is really reaching out, enabling "what the other guy can do".

Phil "We are that other guy, we want that people are able to match with our processors (laugh)."

Bob "So to me, Torrenza is really the spirit of saying: "There are different types of compute that are optimal for specific applications" and we want to make sure that happens."

AMD knew very well how to capture the hearts and minds of Hollywood and unlock the creative studios from the locked out platform of yester year. They are expecting that both IBM Roadrunner project (gluing Cell to the Opteron) and next-generation of Cray architecture (DARPA's HPCS Phase 3) set the ground for peta-flops levels of computing power.

INQ "This question will try to clear the misconceptions that many partners in the industry have about the direction ATI is heading. The thing with GPGPU initiative is opening of the architecture. Now, Nvidia is a company that fully focused its GPGPU initiative on open industry standards (Nv CUMA), using C programming language, OpenEXR, IEEE 754 precision and so on... Is ATI, or now AMD-ATI… DAAMIT!, planning to more open the architecture of the GPU chips so that industry could develop in the same way as Nv plans to do?"

Bob "Do you mean Cg? It is not an open architecture. Because, it you look at the architecture, we believe that both AMD and ATI are at the forefront of openness. But we are only at the beginning. Take a look at GLSL (also known as GLslang), it's an open standard - and then there is Nvidia with Cg, which is not open.

"Couple of points I want to bring up. There may be a misunderstanding that we're using something that is not up to standards. And in truth, what we found is that by working with some people, working through OpenGL or DirectX gives the performance they want. But there was good amount of demand by companies such as PeakStream that said: "I want to maximize every single cycle out of that chip", much like people that used to program in microprocessor assembly language, they wanted that access which really controls the CPU usage. And that's why we offered low-level targeted interface for researchers and bleeding edge technologists, to enable them to do it, to see and get the benefits. But, at the same time, we did nothing to prevent people from getting that functionality through open standards. So, we are not going to low-level microcode access instead of API, but instead offering both to interested parties."

Phil "Opening the architecture on both sides is the exact thing we did with x86-64, or AMD64. We were working with the Microsoft about creating an open standard to extend the ISA, worked with partners… and doing it in any other way just does not make sense. I think everybody is saying that we did a very good job with 64-bit extensions, to do them in a very open way, both Unix and Microsoft community were well aware about what is going on, and this has served us well."

While it may look like Nvidia is now leading the field up with CUMA, the truth is that ATI opened its architecture over a year ago. That's why you now have Folding@Home and Stream Computing infrastructure in place. In order to take a look what will happen with ATI GPU tech, re-read what Phil statement about x86-64 deployment.

INQ "That brings the question of drivers. AMD has been a staunch supporter of Linux, while many users of ATI had a hay-ride with drivers for Linux operating system. Nvidia, as the prime competitor has support of Linux community, while every once a while we hear news about petitions to ATI, drivers not working as intended."

Phil "AMD is driving the industry to an open world, and we focus our strengths and with combined approach, achieve what's best for development of the industry around us. All ISVs are important for us."

INQ "And the final questions: AMD Eco-system. This is something whole industry wants to hear, and if possible, I would like to end the rumour-mongering from forums, at this very spot. Basically, AMD and Nvidia go a long way back, there was an agreement between the two companies named SNAP, the Strategic Nvidia-AMD Partnership, you guys worked on the first Xbox design, then got pushed out by Intel at 11th hour. But then, Nvidia produced Enforce platform and put a commanding foot in the world of chipsets. Also, the upcoming 4x4 system uses Nvidia Nforce 680a chipset… so, how do you actually see the eco-system of AMD? The rumours that AMD is shutting Nvidia out, that there will no longer be high-end products from ATI."

Phil "We are a very open platform and I will give you two examples of that. The PCI Express bus is an industry standard and we are going to keep it that way and if people want Nvidia graphics accelerator, it will work as well on our platform as it will work on an Intel platform. Likewise, we welcome ATI graphics cards that plug into the Intel systems, and we will continue to optimise the performance of ATI products on other platforms, just as well as they run on ours. So, it's not that we are a competition. We want that people play at common standards level and so again, we love to see Nvidia graphics accelerator running on our machine, and we love to see ATI graphics running on an Intel machine. Thing is, we say we want to do it on industry standard interface."

And that's what we are all about. Listening to industry and implementing new standards."

We wish to thank PR crew at ATI and AMD Italy for giving me the opportunity to have a conversation in excess of one hour with two leading technologists of DAAMIT.

This is Part One of five in the series. Part Two will follow tomorrow.