AI
Saturday, May 01, 2004 8:26:59 PM
I don't think it is clear yet
There was a long string of technical posts where we talked about how higher temperatures in general increase the leakage currents of the device. "Normal" operating temperatures are between 50C and 75C depending on how much work the processor is doing. To enter the "Avalanche" processes that Pravin was talking about would require the silicon to be at or above 150C or so, which is very unlikely. Of course a P4 is going to throttle well before it gets anywhere near that temperature. I think throttling starts at 70 or 80C.. as such we don't really need to worry about those mechanisms we were discussing.
In general, you can think of the device power consumption as having two components: active power, and leakage. Active power= constant * frequency * voltage^2. Active power goes down as a fairly strong function of temperature. The leakage power is pretty much a constant, but does go up as a function of temperature. In their ISSCC paper, Intel mentioned that prescott power is about 80% active, and 20% leakage. This leads us to believe that for the 100W that a typical prescott can draw, about 80W fits the CFV^2 equation, while 20W fits the "leakage" portion you have talked about that increases with temperature.
If people have not noticed, there is a "very first" look at the prescott D-0 posted here in russian:
http://www.overclockers.ru/news/webnews.shtml
Not much is covered, but we do notice that the D-0 operating voltage has dropped to 1.375 from 1.400 for the C-0. The reviewer also notes that the device runs much cooler than the C-0, but is still warmer than the Northwood. That is about the extent of the analysis, beyond a relatively meaningless 3.9Ghz overclock result. I am sure we will see more over time.
In terms of process changes between Dothan and Prescott, I seriously doubt there are any. Intel is very loath to run a custom process for anybody.
--Alan
There was a long string of technical posts where we talked about how higher temperatures in general increase the leakage currents of the device. "Normal" operating temperatures are between 50C and 75C depending on how much work the processor is doing. To enter the "Avalanche" processes that Pravin was talking about would require the silicon to be at or above 150C or so, which is very unlikely. Of course a P4 is going to throttle well before it gets anywhere near that temperature. I think throttling starts at 70 or 80C.. as such we don't really need to worry about those mechanisms we were discussing.
In general, you can think of the device power consumption as having two components: active power, and leakage. Active power= constant * frequency * voltage^2. Active power goes down as a fairly strong function of temperature. The leakage power is pretty much a constant, but does go up as a function of temperature. In their ISSCC paper, Intel mentioned that prescott power is about 80% active, and 20% leakage. This leads us to believe that for the 100W that a typical prescott can draw, about 80W fits the CFV^2 equation, while 20W fits the "leakage" portion you have talked about that increases with temperature.
If people have not noticed, there is a "very first" look at the prescott D-0 posted here in russian:
http://www.overclockers.ru/news/webnews.shtml
Not much is covered, but we do notice that the D-0 operating voltage has dropped to 1.375 from 1.400 for the C-0. The reviewer also notes that the device runs much cooler than the C-0, but is still warmer than the Northwood. That is about the extent of the analysis, beyond a relatively meaningless 3.9Ghz overclock result. I am sure we will see more over time.
In terms of process changes between Dothan and Prescott, I seriously doubt there are any. Intel is very loath to run a custom process for anybody.
--Alan
