Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Advanced Micro Devices - Moderated (AMD) -- Ignore unavailable to you. Want to Upgrade?

To: BUGGI-WO who wrote (233100)	5/18/2007 3:24:15 PM
From: pgerassi	Read Replies (2) \| Respond to of 275872

Dear Buggi: The half of the power equation overlooked by many is that power is equal to dynamic power plus static power. The latter is typically referred to as leakage. The equation is thus P = a * f * v^2 * n + b * v * l(v, t) * n where f is the clock rate, v the core voltage, n the number of transistors, t the temperature in kelvin, l() the function showing leakage based on voltage and temperature and a and b showing constants based on circuitry and die layout. Given that leakage is now quite significant, you can't ignore its effects. Its why small decreases in frequency lead to large decreases in power. Leakage exponentially increases with voltage and it reduces with higher temperatures in many cases. So a small change in voltage leads to magnitude changes in leakage current. So while Barcelona optimizes for performance and less for overall power, transistors are chosen more for the speed at which they work. Thus Barcelona idle power has a higher leakage at any given voltage. OTOH Griffin optimizes for reduced leakage because far more of its time is spent idling. Lower leaking transistors don't switch as fast, thus lowering maximum clock. That means that certain critical paths in Barcelona won't be the critical ones for Griffith. This makes the layout different and that means that both go down different paths anyway. This means that you remove less critical features and add more critical ones which come from the different goals. Still many choices will be the same and experience of the other path still has value. Look for Griffith to have things like load/store reordering, sideband stack optimization, 32 byte ifetch, wider L1 L2 paths and other such things that help integer IPC. Virtualization would still be a good thing to have as well and keeps all code portable. Higher IPC helps keep the clock down and that allows more time for idling and lowers average power. Pete