Undervolting a Laptop 262
Delph1 writes "Laptops often comes with two Achilles heels, heat and limited battery time. There are, if not cures, at least remedies to make them less obvious. By lowering the voltage to the processor you can not only drastically lower the heat dissipation, but also increase the battery time significantly. NordicHardware gives a nice walk through on the process and was able to boast 18% lower temperature and a 20% reduced power consumption."
Computer Performance (Score:4, Interesting)
If there are no performance problems, then why dont all laptop manufacturers already do this?
--
No Con's? (Score:4, Interesting)
isn't this what speed step did back with the PIII? (Score:3, Interesting)
"Mobile Intel® Pentium® III processors with Intel SpeedStep® technology let you customize high performance computing on your mobile PC. When the notebook computer is connected to the AC outlet, the new mobile PC runs the most complex business and Internet applications with speed virtually identical to a desktop system. When powered by a battery, the processor drops to a lower frequency (by changing the bus ratios) and voltage, conserving battery life while maintaining a high level of performance. Manual override lets you boost the frequency back to the high frequency when on battery, allowing you to customize performance.?
What's New? (Score:1, Interesting)
Parent is a Bad Idea (Score:3, Interesting)
I'm not sure which is more surprising (Score:3, Interesting)
Too much misinformation here. (Score:5, Interesting)
Undervolting is NOT underclocking. You run the same clock speed, you just provide the CPU with less juice.
You do NOT need to underclock to undervolt, though if you're trying to hit a super-low voltage, a lower clockspeed will let you do it.
It can be perfectly safe. If you undervolt, and successfully run a Prime95 torture test for 24 hours, you're pretty much set. I'm currently running a 1.8Ghz Dothan Thinkpad at 1.134V (default at 1.8 is 1.340), and 0.700v at 600Mhz (default is 0.980 volts). That's on par or lower than those 1.0Ghz ULV Pentium-Ms!
Lowering the voltage does not affect performance (Score:2, Interesting)
Re:Counter productive maybe? (Score:3, Interesting)
You might be removing the ability of the system to manage its own power. This was the case with my desktop. Dropping the CPU frequency on my P4 based desktop actually made it consume more electricity. At its factory speeds, the system uses abotu 90W when not doing a whole lot, and about 215W when under heavy load. Dropping the CPU frequency to 300MHz caused it to idle at about 110W usage. I did not experiment with dropping the voltage however, which may have produced a net savings.
Undervolting is NOT the solution (Score:0, Interesting)
Most circuits in laptops are desgined to their limits of design and manufacturing production so I personally won't screw this up. Desktop or workstations have an little more leeway on design so I would play this theory on those first.
USB often has its corners cut... (Score:3, Interesting)
Just food for thought.
Re:isn't this what speed step did back with the PI (Score:3, Interesting)
This is what AMD did with their PowerNow!(TM) [amd.com] technology. It dynamically adjusts CPU power consumption based on CPU load. According to AMD, it can reduce CPU power at Idle by 75%. I know on my laptop, I can hear the fan speed up and slow down based on the load on the CPU.
Re:What for? (Score:3, Interesting)
But anyways, you do bring up an interesting point. usual power saving features do things like lower the clock rate when not in use, but lowering the maximum clockrate you would lower the speed of the computer, thus the max power it puts out. Knowing that you will be running the processor at max speed longer, you may or may not gain power/heat savings overall for long complex tasks, but I imagine for simple tasks you would.
now - I don't believe you need to downclock just because you lowered the voltage, but you may introduce system instability
That's usually what happens when you overclock without increasing the voltage as well.
Re:Counter productive maybe? (Score:3, Interesting)
Speedstep can only throttle my processor down to 600MHz (from a max of 1.2GHz) but underclocking reduces it to an effective 300MHz.
I do not notice the performance hit, and I do a lot of photo editing on this machine.
--Pat
Re:Undervolting is NOT the solution (Score:3, Interesting)
Resistive loads (which, to a first approximation, a CPU is), don't "demand...Watts", they "draw current". The load resistance doesn't change, so Ohm's Law I=V/R says that if you drop the voltage, the current decreases. Drop the voltage 5%, you DECREASE current 5%. Your total power (V * I) is now decreased by 10%.
As for "screw up the reference voltage", this is and remains system ground, or 0 V. Yes, at some lowered voltage, the CPU will cease to operate. Assuming the CPU still runs, a logic level of 2.85 V is just as good as 3.0 V.
Sacraficing speed for power? (Score:2, Interesting)
Ok all that for 10 min of battery life? (Score:2, Interesting)
Re:Sacraficing speed for power? (Score:2, Interesting)
Please use Kelvin to compare % temperature change (Score:1, Interesting)
> was able to boast 18% lower temperature
FYI: You can only compare temperatures this way if they are measured in Kelvin.
If it was not measured in Kelvin, then you must convert it to Kelvin before computing the percentage.
Even then, it can be risky to compare temperatures in this way, because there is sometimes a non-linear relationship between temperature change and thermal energy change.
(For example, when you heat up water, sometimes the added energy goes mostly toward increasing the molecular kinetic energy, but sometimes the added energy goes mostly toward breaking inter-molecular bonds. Those two phenomena have a very different impact on how much the temperature rises.)
Comparing energy or power levels is usually more meaningful.
Re:Ok all that for 10 min of battery life? (Score:3, Interesting)
Re:Counter productive maybe? (Score:2, Interesting)
Of course many users think they can do better with their over/under voltage/clock regimes, all the while claiming that it's a conspiracy by the processor vendors to deliberately throttle down their CPU. But all they're doing is wiping away the safety margins built in by the engineers. It's kind of like ricers cutting springs in their hondas. Never mind that Honda engineers spend thousands of man hours and million of dollars testing, simulating, and tuning their suspension on their expensive workstations and test tracks. I see their cars bobbing up and down like low riders over small imperfections on the highway and shake just my head.
In short, people who think they're smarter than engineers, usually aren't.
Rovclock (Score:5, Interesting)
While it doesn't actually reduce voltage, it can be used to underclock GPU and memory speed. My somewhat unscientific testing has shown no major differences between fglrx and radeon + rovclock with 2D, but I did note a 27% decrease in battery draw for 3D using the fglrx driver.
Of course, you're trading performance for battery life, and why you'd want to eg, play a 3D game on battery I wouldn't really understand
YMMV
ws
Re:Bad Idea (Score:3, Interesting)
It should probably be mentioned here that the "lowest safe voltage" (if there can be said to be such a thing) is temperature-dependent (a function of the effectiveness of a fan, the density of the air, the load on the regulators, the number of components powered up vs. down...), and can even be data-dependent.
In order for a CPU to work properly, an awful lot of "digital ballet" needs to be happening, billions of times per second, in perfect harmony. Failing a single setup or hold check on a clocked logic gate leads to erroneous bits entering your data and control streams. Quite a few logic blocks, implemented in quite a few different transistor-level CMOS techniques, have different timings depending on what data is flowing through them at the time -- i.e. whether certain bits are zeroes and others ones, or whether the bit transitions from a one to a zero at the same time as an adjacent wire is switching, etc.
What I'm saying, of course, is that if you "spend a lot of time to find out what is the lowest safe voltage", and then (oops) you plug in a USB peripheral that you weren't using before, or (oops) your room heats up a little, or (oops) you use a laptop cooler to get it off your lap a little (setup and hold times, as well as clock and data paths are voltage dependent y'know...) or (oops) you access an unusual data pattern in a program you haven't run in a while, or (oops) you use your laptop on an airplane, your happy little CPU (that you just had to eke out that last little bit of thermal margin from!) starts failing. Quietly, sometimes.
If you're lucky, Windows just crashes -- but manages not to trash your hard drive while doing so (remember, this is weeks or months after you ran your little three-day burn-in marathon "torture test"). But in a not-so-pleasant outcome, your data just starts quietly... rotting. Binary file formats start getting corrupted. Programs stop running. Checksums start failing. Spreadsheets start changing quietly behind your back. I grant you that on a Windows-based laptop this sort of thing could generally be considered the norm (due to virii and spyware and so on), so you might not really notice.
But fortunately for the retarded selves of the people "publishing" (I use the term loosely) this crap, there are thousands of engineers at Intel and AMD running detailed and comprehensive timing simulations of all of the tens of millions of transistors in those teeny-tiny CPUs. They can't possibly EDAC- and parity-protect every net in the design, but they do run static and dynamic timing checks, with probabalistically-developed parasitics and 3-dimensional noise parameters. They consider crosstalk and electrical noise, process variability, and electromigration. They run extensive tests on the CPUs at the wafer and package level to detect slight variations in current consumed during carefully-chosen JTAG test vector execution. They speed-bin and scrap parts that don't pass the multi-billion-dollar test regimens that they spent close to a decade perfecting, and use the detailed results to craft application notes and engineering design guidelines that they feed to their OEM partners, who in turn carefully couple the CPUs to circuit boards with traces perfectly matched to one another, as well as to the impedance characteristics of the solder they're using, and attach them to advanced power supplies and controllable oscillators.
And then you go ahead and download a shitty little ninety-nine cent program off teh unternet and procede to pick your own CPU voltage that "seems to work OK".
Yeah, good luck with that. Sounds like a plan to me. There once was this handy expression about fools and money -- I guess "data" is just the modern substitute.