Taking on ARM with Moorestown

While Menlow uses a more traditional PC setup (cpu + chipset), its successor due out in 2009/2010 won't.  Moorestown, like Menlow, is a two-chip design - composed of a north hub and a south hub.  The north hub features the CPU, memory controller and graphics, while the south hub is just I/O.  This should sound a lot like AMD processors today, but there's one key differentiator: the south hub won't support the PCI bus.

The lack of PCI support is purely a power savings measure; while Intel expects Menlow to be far faster than the ARM platforms used in smartphones today (approximately 2x the performance of what's in the iPhone), ARM will continue to have a significant power consumption advantage. 
The ARM power consumption advantage comes from two places: active power and idle power.  Intel plans to make significant improvements in active power over the coming years, eventually bringing parity with ARM but it will take some time.  These improvements will come via architectural changes as well as improvements in manufacturing processes.  
The idle power reduction will come through highly integrated platforms, like what we're describing with Moorestown.  By getting rid of the PCI bus and replacing it with Intel's own custom low-power interface, Intel hopes to get idle power under control.  The idea is that I/O ports will only be woken up when needed to (similar how the data lines on the Centrino FSB function), and what will result are platforms with multiple days of battery life when playing back music. 

While there will still be some Menlow devices that run Windows, Intel expects all Moorestown devices to run Linux.

Intel expects that in 2 - 3 years, thanks to reduction in die size and thermals, that it will be in a device the size of the iPhone.  Until then we'll have to deal with larger MIDs for Menlow and the candy bar form factor we saw at last year's IDF.
Menlow based MIDs Final Words


View All Comments

  • n0nsense - Sunday, January 13, 2008 - link

    I don't like intel.
    Usually it does something good only when they have no other choice (like Core architecture).
    The x86 is not good from the very beginning decades ago. But greater evil is there. It's name Microsoft. They just want you to use what they designed to get money from you instead of designing For You with user in mind.
    And the only thing that make me smile, is expected percentage of devices based on this platform that will run MS products.
    Actually Linux rules in mobile/efficient devices world.
    As for the software compatibility on different platforms, Java does the job.
    And the Linux world does know how to work with different platforms. Most distributions available for several architectures.
    So, the most important thing is to have size/power/performance in right combination.
    It will be very sad if intel will succeed to kill other architectures.
  • Boissez - Friday, January 11, 2008 - link

    I've just checked the numbers... And I just can't see how this 47M transistor processor clocked slower than the 77M Pentium M can perform similarly.

    I know that there has been som architectural tweaks meanwhile... but still we're talking about a 3x improvement (assuming the 600mhz part=1000 Mhz Pentium M) - If those performance claims are true I'm inpressed.
  • IntelUser2000 - Friday, January 11, 2008 - link

    Actually, most of the differences between the 47 million and 77 million is cache. If you look at the core alone, its not much different. Of course, it will impact performance...

    Pentium 2x8KB=3.1 million
    Pentium Pro 2x8KB L1=5.5 million
    Pentium II 2x16KB L1=7.5 million
    Pentium III 2x16KB L1=9.5 million
    Conroe/Merom 2x32KB L1=19 million
    Silverthorne 32KB iL1, 24KB dL1=10 million??

    Silverthorne's core transistor count is estimated at 10 million. Maybe it can perform like a Pentium III 512KB cache with SSE3. Anyway, it has 512KB L2 cache so...
  • IntelUser2000 - Saturday, January 12, 2008 - link

    The estimated performance of Silverthorne:

    50-60% faster in multi-thread than A110
    Similar in single thread to A110

    A110 MT performance
    A100 ST performance

    (A100=600MHz/512KB/400MHz FSB/Dothan)
    (A110=800MHz/512KB/400MHz FSB/Dothan)

    When Intel says "similar" performance to 2004 notebooks, they are referring to ULVs. ULVs are clocked at 900MHz-1.1GHz. Take a look at Intel's technical documentations for their ULV processors. You'll notice that the performance increase has been almost flat even with the architectural/process improvements. They wouldn't have been able to make a 0.5W CPU without a substantial architectural overhaul.

    Pentium M 130nm ULV
    1.1GHz 7W 400MHz

    Pentium M 90nm ULV
    773 1.3GHz 5.5W 400MHz

    Core Duo ULV
    U2500 1.2GHz 9W 533MHz

    Core Solo ULV
    U1500 1.33GHz 5.5W 533MHz

    Core 2 Duo ULV
    U7600 1.33GHz-10W 533MHz

    Core 2 Solo ULV
    U2200 1.2GHz-5.5W 533Mhz

    For two generations ahead in architecture and process technology, the latest ULV CPU offers much less than it does on desktops and laptops.
  • ajuez - Friday, January 18, 2008 - link

    I think thet the Via C7-M ULV is similar in approach to the Silverthorne processor: very simple architecture focused on power consumption and cost.

    The VIA C7-M ULV consumes 3.5W (according to http://www.via.com.tw/en/products/processors/c7-m_...">http://www.via.com.tw/en/products/processors/c7-m_... )

    However, Intel has the advantage of their manufacturing technology (Silverthorne is a 45nm processor, while C7 is a 90nm processor). I hope Via to update their processor with a better process (55nm?)
  • IntelUser2000 - Wednesday, January 16, 2008 - link

    Maybe we are all confused after Intel introduced the Core 2 Duo. From what I can remember, its only worthy product that came out of the company for many years. All the chipsets after 865/875 were mediocre, and everything else is just a flop.

    I heard that Menlow is headed by the same guy that made a failed XScale and Itanium. Maybe its gonna be another flop. Currently, the only thing that Menlow is sure to bring is the size. Everything else, like battery life and performance is looking gloomy every day.
  • Flunk - Friday, January 11, 2008 - link

    Way less cache, cache is the biggest use of transistors on chips designed for PCs. Cut the cache to say 256k (which is enough for a device running entirely off of flash) and you save a lot of transistors. Reply
  • TA152H - Friday, January 11, 2008 - link

    What's processor cache got to do with Flash drives?

    Processor cache speeds up RAM accesses, RAM cache can be used (and is normally) to cache hard disk accesses, and that's probably what you're thinking of. But they're two different things.

    I agree though, transistors are not the same as size. You can pack memory transistors much more closely than logic, and it can seriously impact transistor count while less seriously increasing the size (and cost) of the processor.
  • iwodo - Friday, January 11, 2008 - link

    If Silverthrone is really that powerful i think it can used as a NAS Chip.

    All current NAS are so underpower and dont provide fast enough transfer due to CPU limitations. Windows Home Server kind of remove the performance barrier since it require Desktop class PC.

    However they are also more power hungry too.

    I hope Apple use this as their base for Apple Home server
  • Shadowmage - Thursday, January 10, 2008 - link


    Nice article, but you spelled "Bonnell" wrong.

    Menlow consists of the Silverthorne processor and the Poulsbo chipset. If you're in dire need of yet another codename, the core used in Silverthorne is also known as Bonnenn.

Log in

Don't have an account? Sign up now