Re: Powerbooks and Variable Timing
Originally posted by Macrumors
PowerPage.org claims that upcoming Powerbook revisions may employ new motherboard technology involving variable timing to extend battery life:
This is presumably distinct from the processor cycling advantages that current laptops already perform to increase battery life.
Powerpage continues to maintain a January release of new Powerbooks.
Not sure about PowerBooks (I only have a Wintel laptop), but Intel/AMD "mobile" chips have employed "speed step" technology for quite some time. "Speed Step" reduces the operating speed of the CPU (by reducing its clock multiplier, leaving the bus speeds constant) to a
fixed low-power setting whenever the BIOS notes that the computer is running on battery power. This happens regardless of CPU activity, and has been known to cause "hiccups" in early implementations of the BIOS (ie, you'd unplug the laptop and it would freeze up for a few seconds or indefinitely, or plugging/unplugging within a certain time interval would cause a BIOS panic and it would shut down).
It's hard to judge from such a brief rumor, but it sounds like the upcoming PB technology is more akin to what Intel is aiming to do with its Banias processor, although the addition of reducing bus speeds is, I think, beyond Intel's current plans. The "variable" nature, if accurate, is also an advantage over Intel, which is "fixed" in speeds (ie, Intel SpeedStep is like a three-way lamp except the "medium" setting doesn't exist: it's either on "high" or it's on "low" or it's "off"; "variable" timings would be more akin to a wall dimmer switch with a relatively "infinite" number of settings between "high" and "low".) This would allow the laptop to more appropriately use power in the middling situations where "low" speed isn't enough processing power for the app, but "high" speed isn't necessary. Banias technology does retain the advantage, however, of being able to purportedly switch off sections of the CPU that are not in use, which would be like taking power from the AltiVec unit when not in use or the FPU, and only restoring it when an instruction that uses that unit shows up in the pipeline.
CPU load-based power cycling is a neat trick, but also relies on the fact that software has to be able to tell the OS and hence the BIOS if it is "really" using all those CPU cycles for a user-dictated reason (ie, rendering a frame of a movie) or if it is just "taking advantage of" free cycles to do its processing and has no user mandate to get its work done quickly (ie, a scheduled disk indexing operation or automatic virus scan, etc). This could be done on first approximation basis according to the app's threading priority, but that's really not quite the same thing.
Imagine working in your favorite word processor: 99% of the time, "low power" mode gives more cycles than your fingers can keep up with, but when you go to write the thing to disk or do a full-text search on your 700-page novel, you'll want all the cycles the CPU can give you. This is something that Intel/AMD technologies can't do right now, and that the rumor would appear to indicate PowerBooks would handle properly.
So, if January is really the date for such tech to be in PowerBooks, the PB camp will have a HUGE advantage over Wintel notebooks (not that PowerBooks were lagging ...)
Also, regarding OS X and power usage: OS X tries to keep all subsystems active and doing "work" at all times. This allows a whole heck of a lot more "eye candy" and behind-the-scenes niceties to occur than would have been possible in OS 9 or below. By "behind the scenes niceties" I mean not only automated tasks, but also things like prioritizing memory and disk space usage and intelligently managing the preemptive multitasker and keeping memory accesses in check, etc. These are not things that can just be "turned off" easily, as the entire system relies upon these things being done and done properly.
And, yes, you will find that most Unixes are not really the best OS's for mobile machines. Unix just provides too many services to thrive on a battery-powered device. On the other hand, you can fashion Unixes which do well in such situations (like the Linux base of Zaurus handhelds), but that means giving up many underlying services and sometimes some of the overall stability for which Unix is famous.
Note that on my Wintel (HP) laptop, Win Me (just a half step up from OS 9 in terms of complexity) would run (doing "nothing") for about two hours, Win XP runs for an hour and a half, and Linux runs for about an hour and fifteen. Granted, neither XP nor Linux are "tuned" highly to get rid of "sometimes needed" services/daemons, but that's how the battery usages pan out.
