AMD's big bet on Llano CPU+GPU

AMD's big bet on Llano CPU+GPU

Summary: Intel may own the present, but AMD believes the future belongs to Fusion. At a recent conference AMD provided the first glimpse of its first Fusion APU slated to ship in 2011.


Last week I was at ISSCC, an annual chip conference in San Francisco. Intel tends to dominate these conferences, not only because it is the largest semiconductor company in the world, but also because it is the only one shipping a 32nm microprocessor--and still the only one manufacturing chips using high-k material and metal gates in high-volume. (Earlier I previewed Intel's Westmere presentation at ISSCC.) While Intel may own the present, AMD believes the future belongs to Fusion. And at the conference AMD provided the first glimpse of some of the components that will go into its first Fusion APU, or Accelerated Processing Unit, slated to ship in 2011.

The idea of an APU is to combine an x86 processor with a capable graphics processing unit (GPU) on a single silicon die. AMD argues that simply adding more and more general-purpose x86 cores to a chip will yield diminishing returns since many--though certainly not all--tasks can be handled more efficiently by a highly-parallel processor with lots of little cores such as a GPU. AMD happens to have a capable DirectX 11 GPU, the ATI Radeon 5000 series.

This is not a new concept. It's the same idea behind GPU computing, but unlike Nvidia, AMD is in a position to physically combine the CPU and GPU. Intel already sells processors with a graphics controller--its Core i3 and Core i5 dual-core laptop and desktop chips--but it's not a single chip. Rather Intel combines a 32nm CPU with a 45nm GPU into one multi-chip package. Like AMD, Intel plans to put both on the same physical die, but the timing is unclear.

AMD first APU, code-named Llano, will also be the company's first processor manufactured at 32nm using high-k and metal gates. Llano will consist of four x86 cores, a DirectX 11 GPU and a DDR3 memory controller. It will be part of Sabine platform for desktop replacements and thin-and-light laptops. AMD will have a separate platform (Brazos) for ultraportables and netbooks with a new x86 design known Bobcat. This is also scheduled to ship in 2011, but AMD hasn't announced what process node will be used. Llano will also be used in mainstream desktops while another new core, Bulldozer, will show up in four- and eight-core processors for high-end desktops. Both will be 32nm processors and will be available in 2011.

Given all of the changes in Llano-a new process node and new materials on the first monolithic CPU and GPU--it's little wonder that AMD has largely stuck with the same design for the x86 as the current 45nm cores. AMD isn't ready to discuss the entire APU; instead it focused on these cores, each of which contains about 35 million transistors and occupies about 10 square millimeters (not counting the 1MB of L2 cache). Of course when you combine four of those with lots of cache, a GPU, memory controller and peripheral circuitry, the final APU will be much larger. AMD has previously said it will have around 1 billion total transistors at 32nm.

Most of the changes to the core have to do with better power management. For example, individual cores can be completely disconnected from the power supply when not in use to minimize power. The cores in Llano will operate at 3GHz or faster while using anywhere from 2.5 to 25 watts. AMD also talked about some of the changes in the circuit design required to make the design work with the high-k dielectric and metal gates. But the bigger questions surrounding the integration of the GPU remain a mystery for now.

From a user standpoint all that really matters is whether AMD can successfully deliver Llano on time. There's no doubt Llano represents a big technical challenge with a new process technology, new materials, and what looks like the first on-die GPU--all produced under a new manufacturing arrangement with the spin-off of GlobalFoundries. But the AMD has been talking about Fusion since it acquired ATI in 2006, and with Intel already shipping 32nm processors for the same mainstream laptops, AMD can't get there soon enough. If Llano arrives on-time, and it improves performance per watt and demonstrates the benefits of combining x86 cores with a powerful GPU, it should do a lot to make AMD more competitive.

Topics: Hardware, Intel, Processors

Kick off your day with ZDNet's daily email newsletter. It's the freshest tech news and opinion, served hot. Get it.


Log in or register to join the discussion
  • Poor AMD

    I feel sorry for AMD. The idea that the future
    belongs to heterogeneous multicore processors is
    pure folly. Neither the Fusion project nor
    Intel's Larrabee will come to anything. Why?
    Because no new processor is going to succeed in
    this century unless it solves the parallel
    programming crisis. The problem with hybrid
    processors is that they exacerbate the
    difficulty of parallel programming.

    The sad reality is that neither Intel nor AMD
    are innovating anymore. Their management
    consists mostly of baby boomers who are too old
    and set in their ways. Intel is too big and too
    unwieldy and AMD has major cash problems. Worse
    is that another big chip failure will be the end
    of AMD. Intel can always ride out the storm when
    some other company comes out with a powerful
    solution to the crisis and blows everybody out
    of the water. Intel will buy them up, unless, of
    course, the other company happens to be a big
    competing entity like Google, HP or some
    Chinese, Indian or Japanese entity with
    government backing. In which case, Intel is
    toast. Not even giants are invincible in this

    Google "How to Solve the Parallel Programming
    Crisis" if you're interested in the future of
    parallel processing.
    • Wanna bet?

      Actually great post Mapou. However, a bit negative?

      Whatever the situation, I wish luck to AMD and Intel, for they have transformed the landscape of computing as we know it today.

      And nobody, especially NOT the governments can do better than any private sector business IMHO.

      Governments are um, governed by politicians; mainly hypocrites, that will brown-nose one moment, and kick ass the next. Constantly flip-flopping.

      But I will Google "How to Solve the Parallel Programming Crisis" for some enlightening reading.

      Thanks to feedback; I learned something new today.

      A friend to all is a friend to none.
      ~ Aristotle
      (Perfect definition of politicians)
      • RE: AMD's big bet on Llano CPU GPU

        I don't remember anyone mentioning the government.
    • Define innovation!

      They certainly have researched new technologies to keep moore's law running. If you mean that the basic turing engine idea hasn't change well they never really did innovate. In one sense today's new chips are no different than the old 4004.

      In another sense they are radically different on some many fronts that it is hard to enumerate.

      Personally I think Intel's latest chips since they adopted the tick tock approach incorporate way more advances per family than ever before. That is there are more changes going from core 2 duo to core iX families than there were from 386 to 486 to Pentium I - 4

      The parallel problem is a problem for desktops but not as much for servers.

      For me my core i7 combined with Visual Studio 2008 (which allows parallel compilation) allows me to compile nearly a million line of code app in 3 minutes compared to 20 minutes for my hyperthreaded single core. The processor (showing 8 cores. 4 full and 4 partial) ran between 60-70% utilization with all cores showing about equal loading. Toward the end that trailed off because there wasn't enough parallel tasks to keep it busy. The VS solution had dozens of projects (dlls) most of which were not dependent on each other. Also I imagine I/O was a bit of an issue.

      And man this thing cranks on distributed computing projects hosted by BOINC. I love doing protein folding and this thing churns it out fast.
      • RE: AMD's big bet on Llano CPU GPU

        Well, Hypertheading is a gimmick, you're still bound by I/O and Intel copied AMD's innovations and managed to use their monopoly position to thwart superior technology until they could copy it.
  • RE: AMD's big bet on Llano CPU GPU

    This will have Intel worried and could be the first successful attack on Intel bread and butter, that is office computers. It might even have effect on user interfaces on global scale, because office machines with low quality Intel graphics has been holding UI off. Other battlefronts are Media PCs and cheap game machines that are better than PS3 and Xbox 360 (or Media PC and games machine connected to the TV rolled into one).
    And, cheap laptops that can play demanding games.
  • Not as versed on this as others Question

    If the gpu is combined with the cpu will this mean you cant upgrade as you can now like a regulkar video card? As the video cards become better and better people upgarde but not as often with thier processors as the prices would be insane if to upgrade your graphics you had to buy a new motherboard and new matching cpu/gpu just to get better graphics or added feature of like HD tv tuners..or would this be different thing meaning you would also still have a graphics card to use
    • Yes but not in the large and growing laptop/netbook segments

      no one upgrades either in these as it's cheap enough to completely replace them every 2-3 years...
      Johnny Vegas
    • RE: AMD's big bet on Llano CPU GPU

      I think the use of the GPU on the die serves as a math machine extraodinaire. However, the data transfer to the GPU would now happen at the speed of the internal Hypertransport BUS rather than the external limitations of PCI-X. The speed difference for graphics processing should be significant. Hypertransport is much faster than anything Intel has come up with. Couple that with 4 memory controllers and the speed and power increases that you get with 32nm and you've got a killer machine.

      Oh, and you can still add an external video card if you like, but it will be slower.
  • AMD's biggest issue is promices it can't deliver.

    Not picking on them but the reality is, they can do nothing beyound what the custom house is willing to do or spen moeny on now that AMD is gettign rid of their Fabs.
  • This will never... compared like a complete and full GPU (video card).

    Btw, Intel is doing the very same thing!

    Fusion got fusioned?
    • RE:This will never

      The fact is that Intel has nothing comparable. Intel graphics solution are far far less powerful than Llani
  • RE: AMD's big bet on Llano CPU GPU

    a CPU/GPU combo will have to go in office style desktops, netbooks, and laptops and will dominate those markets, given time, and proper pricing.

    but a CPU/GPU combo in a standard desktop will not sell very well to anyone under ~35 as most people under ~35 are more willing to upgrade (or have their friends) upgrade certain components when they feel they have out grown them, instead of buying a complete new system. Therefore they will want a separate GPU card to put in their PC.

    The combined CPU/GPU system will create a new-ish segment of the market, for those that plan on buying, then keeping it until it stops powering on.

    I have no problem with this, and think its a great idea, if heat disipation can be controlled in these smaller devices.

    for the desktops, picture the nano/micro-ATX style PC's that handle specific jobs or minimal usages, give these a combined core sporting quad core CPU and multiple GPU's, that will make a lot of people happy.
  • RE: AMD's big bet on Llano CPU GPU

    I think we are missing the main point here. The big money is not in the high end technology and neither it is in the low end technology. The big money is in the main stream technology and that is a big lesson that AMD has tough Nvidia. So AMD is going after the mainstream market with Llano: laptops and desktops for the value market. And clearly Intel is not ready for that (otherwise where are the 32nm quad cores from Intel? I guess they are not in the market for a reason...). So AMD is going to play with Intel a similar game he played with Nvidia: maintream customer don't care about how many GFLOPS your CPU can make they care about price and about what can you do with the computer you buy... and AMD lately is good at listening what customer want while Intel is overcharging for their technology. PC manufacturers I am sure that are tired of Intel commercial practices, so if AMD shows where the money is with Llano I think the share of the value market can become closer to 50% for AMD.