Slashdot Mirror
AMD — "We're Not Entirely Honest" About Batteries
Slatterz writes "In an apparent attack of the bleeding-obvious, an AMD rep has come clean and admitted (on behalf of the industry) that notebook and phone battery life figures are completely unreliable. AMD's senior vice president Nigel Dessau says that 'we are not being entirely honest with users about what PC battery life they can expect to actually experience.' He says AMD will now use a combination of idle time (where the machine is left to sit idle, and timed to see how long it takes for the battery to go dead), and 3DMark06 to measure battery life. Great in theory but some of the industry already bases battery figures on a two-test measurement, and the results are still wildly inaccurate."
No, that's the thing. Everything they've told you is technically true... under certain conditions. Possibly even the conditions that they've listed in a small-print disclaimer (available upon request, if you can arm-wrestle the tiger and win).
The World Wide Web is dying. Soon, we shall have only the Internet.
If you recall, AMD's performance rating was an important step forward for the CPU manufacturer industry at that time. Intel was pushing for higher and higher clock frequencies with longer and longer pipelines - something that made little sense.
Performance ratings allowed consumers to effectively compare AMD and Intel chips side by side in ways that are useful.
You can't legislate goodness. Let each to his own destiny, by will of his freely made choices.
3) Advertise "minimum" battery life
What is wrong with that? Then I can expect at least 40 minutes of battery life and anything more than that is nice.
What's wrong with that? What's wrong is that you're telling the customer a number that by and large they aren't interested in. What they want to know is if they can watch a full DVD without recharging. If they can work on their Excel spreadsheet for the entire 6-hour cross-country flight if the plane doesn't have plugs. You tell them "minimum 40 minutes" and they say "Whoa! That's not long enough to do anything!" and you say "Well it's just the minimum, under typical usage conditions it will last much longer," and then they ask "And how long is typical? Long enough to watch Casino Royale on BluRay?"
What's your answer? Hypothetically you should be able to actually say whether it'll last long enough to watch the movie, but how do you answer that question in general? What is "typical"? That's what people really want to know, the minimum number doesn't really do them much good except to say that if they really load down the laptop, it won't last long. Which makes the product look bad, and is still by and large not that helpful.
It's not an easy question, more difficult in many ways than talking about performance. Considering that power has only become a major concern for commodity chip makers in recent times, I'm not surprised that their battery life estimates aren't very accurate. Of course, whatever estimate they do use, no matter how accurate, will be measured in a way that makes their parts look good. That won't change, ever. I'm sure that's part of your motivation for the minimum time metric -- there are far fewer ways to screw with it. Which is nice, but not sufficient by itself.
The enemies of Democracy are
>>This happens in every industry
This is a bit different from a breakfast cereal saying "now even tastier" or a soap promising "more suds!" The first is subjective (personal preference) but the second is objective -- it can be quantified and proven/disproven.
In this case with batteries, rather than taking an actual measurement of performance, the industry is building an estimate from a combination of measured behavior + a calculation based on a performance variable. It's no different than the automobile industry stating "EPA Estimated MPG city/highway" which is not based on a dynamometer test or actual performance measurement but instead is calculated based on the amount of CO2 which exits the exhaust pipe of the car! Is it any wonder, then, that hybrid cars which shut off their gasoline engine when stopped and at low speed/light acceleration, would give grossly inflated figures? Well, they did (and do), which explains why real-world MPG is often far less than this calculated (not even simulated) performance.
In short, they're both lying and it's obvious. Yet companies wonder why consumers are so cynical and therefore difficult to reach with advertising.
What is needed is real-world testing -- dynamometer ("rolling test track") testing for autos where the wind resistance, temperature, barometric pressure, etc. can all be carefully controlled. Similarly with computers, a pure performance-based measurement is needed which should account for idle time, network activity, etc. Just as an automobile is not tested at full-throttle for 3 hours, neither should a PC, but instead a variety of benchmarks (gaming, web browsing, spreadsheet, word processing, ???) could show performance figures for various activities.
In short, manufacturers, we want real numbers free of hype.
"We're Not Entirely Honest" = We've been lying
Actually, "We're Not Entirely Honest" = "We have no idea how to give you an accurate estimate." As someone whose found himself sucked into the battery/mobile power side of a project recently, I can understand why they'd face difficulties.
When it comes to batteries, there are really only three options for measuring how much power is stored: completely drain it over several cycles to see what you get (which is how the manufacturer confirms capacity, but isn't too useful in situ); test the voltage across the terminals and estimate absed on pre-measured battery curves (which is difficult because voltages don't change dramatically until they're nearly drained); or, in some chemistries, measure the temperature changes in the battery (which detects inreactions that don't happen until the battery is almost completely drained). In practice, all you can do is take the manufacturer-specified capacity, derate that based on conditions in your application, and test to see if you came close.
In general, pulling more current from a battery disproportionately decreases remaining capacity. In general, it's pretty difficult to respond to sudden surges and lulls in power consumption for a user's unknown power cycle needs without making your estimate jump all over the place. In general, the problem is just a pain in the neck. It's like ordering a margarita with margarita-flavored ice cubes from a waiter whose never seen you before, then demanding to know exactly how long before you'll need to refill it (regardless of whether you intended to chug it or nurse it).
I'm no expert, but you don't have to be to see it's not a trivial problem.
1) If a customer doesn't know how much load playing a DVD is, they don't care about advertised battery life.
O_o Seriously?
You think knowing the % CPU utilization of watching a DVD (or BluRay, if this makes you feel better) is a pre-requisite to caring about the question "Will my laptop be able to play a full DVD between recharges?"
And I suppose anyone who doesn't know engine timings and torque curves wouldn't care about the question "Can I get to Grandma's house without refueling" too. That's weird, because I care about MPG but know very little about the engine physics that inform it. Should they scrap the "city/highway" MPG usage models, and instead tell you what MPG you'd get with the accelerator floored the entire way?
You're being ridiculous. Obviously people will care about being able to do the things they want to do without knowing exactly how much load on the system that actually entails.
Of course, even if I accept this premise, you're still not giving them the information they want. Okay, so I know that the fully loaded laptop life time is 40 minutes, and I know that my DVD player uses 5% of my processor. Now what? What's the scaling factor so I can do the math? Oh right it's not that easy, even if you're an electrical engineer. I know what it is you said I should know, and you still can't answer the question I care about.
2) Minimum is a more useful number because it always applies. Typical usage figures can be plucked out of thin air because it varies too much.
Easy to figure out is not the same as useful. Minimum is rarely useful because it rarely applies to what the person is actually doing. When the "typical" numbers can be 6x higher than "minimum", and what people really care about is "typical" for all the difficulty of defining what that means, then no minimum isn't that useful.
Sure the minimum should be specified. That does not get you out of the tricky problem of estimating 'typical' battery life, because that is what the customer wants to know, and for good reason. If you refuse to give them anything more useful than minimum, then you lose sales because you can't or won't answer the questions they care about.
The enemies of Democracy are