Multiple PLLs wouldn't work, but DLL jitter is no problem

Article: Adaptive Clocking in AMD's Steamroller
By: Robyn Henry (foo.delete@this.example.com), May 13, 2014 11:58 pm
Room: Moderated Discussions
Rob Thorpe (rthorpe.delete@this.robertthorpeconsulting.com) on May 12, 2014 5:42 am wrote:
> I find it interesting that they use a DLL and a PLL to vary the clock speed. I think a better
> approach would be to have several PLLs constantly locked to different frequencies and to switch
> between them. Using an extra DLL in front of the PLL will add noise to the clock.

Multiple PLLs wouldn't be in phase; you'd need a whole additional layer of circuitry to avoid runt pulses when you switch. Which would require synchronization delay to avoid metastability. The goal is to stretch *this* cycle, right now. And a few more that follow, then switch back to the main clock when power has stabilized.

And jitter doesn't matter if they just include that in the stretch; the goal is to have a minimum cycle length, not an exact one. The only time you need minimum jitter is during minimum cycle time operation, and you can have a special "pass-through" output from the DLL for that.

Suppose we decide we need a 10% cycle stretch. Then the clock edges used go 1, 2, 3, 4.1, 5.2, 6.3, 7.4, etc. until things have stabilized. Then after clock 16.2, we decide to go back to regular clocking, so waste the remaining 0.8 cycle and jump to cycle 18, 19, 20, etc.

If there's 10% cycle jitter in the DLL, then you instead do a 20% cycle stretch. 1, 2, 3, 4.2, 5.4, 6.6, and if you have the same 12 stretched cycles, that leaves you on cycle 17.4. So waste 0.6 of a cycle and jump to 19, 20, 21.
< Previous Post in Thread 
TopicPosted ByDate
New article: Adaptive Clocking in AMD’s SteamrollerDavid Kanter2014/05/06 12:33 PM
  New article: Adaptive Clocking in AMD’s Steamrolleritsmydamnation2014/05/06 10:12 PM
  New article: Adaptive Clocking in AMD’s SteamrollerXi Yang2014/05/06 11:08 PM
  Decoupling capacitors on GPUsGabriele Svelto2014/05/08 08:24 AM
    Decoupling capacitors on GPUsBen2014/08/27 10:56 PM
  New article: Adaptive Clocking in AMD’s SteamrollerPoindexter2014/05/10 05:06 PM
  New article: Adaptive Clocking in AMD’s SteamrollerRob Thorpe2014/05/12 04:42 AM
    New article: Adaptive Clocking in AMD’s SteamrollerDavid Hess2014/05/12 09:02 AM
      New article: Adaptive Clocking in AMD’s SteamrollerRob Thorpe2014/05/12 09:46 AM
    New article: Adaptive Clocking in AMD’s SteamrollerRicardo B2014/05/13 02:15 AM
      New article: Adaptive Clocking in AMD’s SteamrollerRob Thorpe2014/05/13 05:30 AM
    Multiple PLLs wouldn't work, but DLL jitter is no problemRobyn Henry2014/05/13 11:58 PM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell green?