ISSCC 2007 Coverage Online

Article: ISSCC 2007: A Brief Survey
By: David Kanter (dkanter.delete@this.realworldtech.com), March 26, 2007 10:28 am
Room: Moderated Discussions
Anonymous1 (anon@anon.anon) on 3/26/07 wrote:
---------------------------
>David Kanter (dkanter@realworldtech.com) on 3/26/07 wrote:
>---------------------------
>>Good morning to everyone,
>>
>>I hope you are all doing well. I have just put up my first article on ISSCC 2007...which
>>is admittedly a little delayed.
>>
>>http://realworldtech.com/page.cfm?ArticleID=RWT032607001221&p=1
>>
>>In this first installment, I survey four different papers, with topics that should
>>be somewhat familiar to regular readers. Three of these papers came from the MPU
>>session, one from PA Semi, one from Sun and one from Intel.
>>
>>There will be further articles on chips and technologies presented at ISSCC, but
>>they will be in a slightly different format. The later articles will be focused
>>on a single processor, with ISSCC coverage as a part of the larger article.
>>
>>In the mean time, enjoy the reading.
>>
>>
>>David Kanter

>Thanks for the article. Is PA Semi able to vary the >voltage in different domains
>on the same chip independently?

I believe so.

>How is this different from >Intel's Foxton? From
>what I understand, Foxton will require no software >interaction, whereas PA Semi's chip does. Is this correct?

Yes, you are correct that PA uses software intervention.

Let me try and elaborate on PA's voltage and frequency modification scheme, and Foxton as well.

Here's my understanding of Foxton:

At a high level, Foxton measures temperature and then adjusts frequency in a relatively narrow band (+/- 20%) to get as close as possible to the TDP.

Foxton measures the on-die power draw, voltage, clock distribution etc. Based on these measurements, Foxton can adjust the voltage AND frequency up or down, dynamically.

PA's part works quite differently. The frequency is determined by software, and AFAIK, there are only a couple of different available options (2,1.5,1,0.5GHz). However, based on the selected frequency, each core dynamically adjusts the voltage to the minimum level that can support operation at the target frequency. You don't have to measure the current at all, since you don't care about temperature, but rather the voltage and the quality of the signals.

If I have made any mistakes in my characterizations perhaps some of our lurkers can comment on the two different techniques.

DK
< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
ISSCC 2007 Coverage OnlineDavid Kanter2007/03/25 11:39 PM
  ISSCC 2007 Coverage OnlineAnonymous12007/03/26 08:16 AM
    ISSCC 2007 Coverage OnlineDavid Kanter2007/03/26 10:28 AM
    ISSCC 2007 Coverage OnlineDavid W. Hess2007/03/26 10:37 AM
      ISSCC 2007 Coverage OnlineAnonymous12007/03/26 11:01 AM
        ISSCC 2007 Coverage OnlineDavid Kanter2007/03/26 11:26 AM
          Looks like Penryn will get single-threaded frequency boostAnonymous12007/03/28 11:03 AM
            Looks like Penryn will get single-threaded frequency boostIntelUser20002007/03/28 10:47 PM
      ISSCC 2007 Coverage OnlineJosé Javier Zarate2007/03/27 04:33 AM
      ISSCC 2007 Coverage Onlineanon2007/03/27 08:57 AM
        ISSCC 2007 Coverage OnlineDavid W. Hess2007/03/27 10:08 AM
          ISSCC 2007 Coverage Onlinesavantu2007/03/27 02:37 PM
            ISSCC 2007 Coverage OnlineDavid Kanter2007/03/27 02:42 PM
        ISSCC 2007 Coverage OnlineDavid Kanter2007/03/27 11:30 AM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell green?