IDF kickoff: Going after AMD, and beyond

  • Beijing (China) – Intel has opened its Spring IDF in Beijing with a flood of announcements. The company revealed first performance numbers of its upcoming 45 nm Penryn processors, half a dozen technology projects including a floating point accelerator, an update to its 80-core research processor and a new enthusiast PC platform that will house two quad-core processors. Can you spot the AMD angle?

    Intel senior vice presidents Pat Gelsinger and Eric Kim as well as chief technology officer Justin Rattner provided an outlook of what to expect from the company this year, next year and within the foreseeable future. In a much more compact keynote than in previous years, large parts of the speeches
    and the products it will enable.

    Of course, there is the 45 nm Penryn core, a new processor that may debut late this year or early in 2008 and that will be offered as a mobile version as well as desktop and server derivates. For the first time, Intel revealed a few performance numbers as well as at least one system spec. When compared to a 2.93 GHz Kentsfield QX6800 CPU, a 3.33 GHz Yorkfield processor (Penryn quad-core for desktops) with 12 MB L2 cache and FSB1333 will gain about 15% in imaging-related applications, 25% for 3-D rendering, more than 40% for gaming and more than 40% faster video encoding with SSE4 optimized video encoders, Intel promises.

    On the server and workstation side, Intel expects its 45 nm FSB1600 processors to deliver up to 45% more speed for “bandwidth intensive” applications, and a 25% acceleration for servers using Java, when compared to a current X5355 quad-core processor. These numbers look awfully close to what AMD has promised for its Barcelona quad-core CPUs and, if Intel will keep its promise, AMD may have to upgrade its initial Barcelona processors not too long after their initial launch in order to hold the pace.

    Due in Q3 of this year is also Intel’s Caneland platform with Tigerton dual- and quad-core processors. The CPUs will replace the aging Xeon MP 7100 series and will not only complement Intel’s 65 nm Core architecture deployment, but also allow Intel to regain market shares it has lost to AMD in the lucrative MP server segment.

    By the end of the year, Intel will also be offering a dual-socket consumer PC platform. Depending on your view, this could either be seen as a reaction to AMD’s dual-socket QuadFX (4x4) platform or as a logical step to offer what some enthusiasts have been doing for some time - to take a Xeon server board and use Woodcrest or Clovertown processors to build a system with two dual- or quad-core processors. Intel calls this project “Skulltrail,” which will support two Core processors with up to eight cores. The platform will also run up to four graphics cards.
    A bid deal in Intel’s future plans apparently are system-on-chip (SoC) designs. Gelsinger outlines “Tolapai,” which is positioned as a enterprise-class SoC that is expected to debut in 2008 and reduce the chip footprint size by up to 45% and power consumption by approximately 20% compared to a standard four-chip design. Tolapai will include a new technology called “QuickAssist”, which is reminiscent of AMD’s Torrenza initiative: Intel plans to combine its multi-core processors and third party accelerators working in Intel-based servers, and develop new integrated accelerators inside the processor itself. The approach includes a software layer to exploit the acceleration potential.

    Eric Kim also had a few SoC designs to discuss, including a chip that will aim to improve audio/video, graphics as well as Internet capabilities consumer electronics devices. An example for such a chip is the CE 2110 SoC, which is based on a 1 GHz Xscale core.

    IDF typically provides an outlook of some research projects that show what the company is working on for next-generation computers. Justin Rattner used the opportunity to show an enhanced version of its 80-core research processor, which achieved 1.8 TFlops in February and now tops 2 TFlops.

    While this chip is purely a research project and is not aimed for production, Intel had some details about another floating point enhancing engine. What AMD will try to do with the help of ATI graphics cards, Intel will aim to accomplish with its own IA technology: Following the trend of massively parallel programming, the company mentioned a new processor architecture codenamed “Larrabee.” While the company declined to comment on details, Larrabee is described to be easily programmable by using “many existing software tools.” Rattner said that the technology can scale into the Teraflop range of performance and won’t target consumer applications: The company will look attract, similar to AMD’s stream processing approach, customers in scientific computing, recognition, mining, synthesis, visualization, financial analytics and health applications.
    We can’t help to note that there are many projects that look all too familiar. Take the dual-socket enthusiast platform, take QuickAssist, take Larrabee and take the Nehalem processor with integrated graphics and there is a boatload of similarities with what we have been hearing from AMD over the past year. Sure, it takes time to develop these technologies and Intel may have had the ideas for some of these technologies way before AMD, but it surely looks like Intel is following AMD on key technologies. In the end, it may not matter who had it first. AMD suddenly does not look so different anymore and Intel’s enormous resources may allow the company to bring some technologies earlier to market than its rival.

    Time will tell the ATI technologies are good enough to allow AMD to take and maintain the lead in certain market segments.