Apple has announced, via Twitter, that it’s working with Consumer Reports to nail down what happened to its MacBook and MacBook Pro hardware. The announcement came from Phil Schiller, who reports the company is “Working with CR to understand their battery tests. Results do not match our extensive lab tests or field data.”
There’s a great deal of smoke around this issue (everyone being off for the holidays probably hasn’t helped), but not much in the way of new data just yet. What we know is that some Apple buyers have reported abnormally low battery life ever since the new MacBooks launched. At first, these issues seemed like they might be limited to the 15-inch MacBook Pros, which use AMD discrete GPUs, but later testing has shown this is not the case. Consumer Reports is not the only publication to report that Apple’s battery life: 9to5 Mac reported relatively low figures for their own tests (6-8 hours).
There are several intersecting issues that, I think, are collectively feeding the issue here. First, there’s the fact that Apple’s battery life estimates are typically extremely good. No manufacturer is 100% accurate, because use models change so much between people, but Apple has a reputation for delivering most of what it promises. For Apple systems to be dropping so dramatically (and even 8 hours when Apple says 10 hours isn’t good), it implies that something else is going on here. Second, there’s the sheer erratic nature of Consumer Reports’ results. If Apple or Dell says that a laptop should get 10 hours of battery life and you regularly get 6, it implies that the company’s battery estimate methodology is poor. If your laptop lasts two hours in one test and 9 hours the next, all while you’re performing exactly the same amount of work, it implies something altogether different.
Third, there’s the fact that modern power management has gotten exceedingly complex. Some of you likely remember when Intel’s SpeedStep technology debuted. The idea of a CPU that could self-adjust its clock in response to workloads and the need to conserve battery life was a huge boost for laptop manufacturers and consumers. Over time, power management technology has advanced considerably; AMD’s implementation of Adaptive Voltage and Frequency Scaling (AVFS) system uses hundreds of sensors embedded around the SoC to gather real-time information about the SoC’s temperature and power states, and adjusts the chip accordingly. Intel hasn’t adopted AVFS for its own chips, but it has added new technologies like SpeedShift to allow its CPUs to move in and out of idle states more quickly and to adjust clock speed dynamically. But all of these changes and technologies come with their own complexities. Some are entirely transparent to the operating system, while others require explicit OS support.
Fourth, as CPU improvements have largely stalled, software, not process node improvements, is increasingly responsible for power savings. This, in turn, makes it more likely that software can screw up. If Intel releases a new CPU that uses 30% less power than the previous generation, that gain should show up in every test case. Now, imagine that Intel uses a combination of better codec offload, GPU processing, and more aggressive CPU frequency scaling to cut power consumption. Suddenly, gains that were previously due to improvements to underlying technology are highly dependent on what the end-user is doing. If you compare battery life in video decode scenarios, the results may look great. Switch to a CPU-centric workload that doesn’t give the chip any time to drop into a lower power-state, and they suddenly show no improvement at all. In fact, the newer CPU may use even more power than the old one if it has a higher max clock.
There are a number of odd things about the Consumer Reports’ results. The variation is extremely strange. The fact that Chrome is reported as returning much better results than Safari is strange. It’s possible that this issue is related to an extremely specific bug or errata that causes a problem in Safari that results in the CPU sometimes not dropping into sleep states properly, or pegs GPU usage where it shouldn’t, or that some particular issue related to GPU acceleration in web pages is causing problems. Apple’s decision to remove the “Time Remaining” metric from its laptops could be evidence that the company knows it has a problem and wants to reduce consumer complaints while it deals with it — or it could be completely unrelated.