Ever since Apple’s iPhone 6s and 6s Plus launched, we’ve seen reports from users who were suffering lower-than-expected battery life or found that the home button became too hot to touch in certain situations. Some of these issues are expected with any new device launch, particularly given that Apple chose to source its A9 SoC from two different vendors — Samsung and TSMC. Samsung had been shipping 14nm hardware longer than TSMC and has already been confirmed as having a smaller die size,
Multiple user investigations have uncovered some significant differences between the two chips, however. Not only do the Samsung-equipped devices reach a higher temperature (reportedly up to 40C as compared to 37C for the TSMC hardware). I’ve been looking over the results — one benchmark run has compared a Javascript loop, while the other ran AnTuTu repeatedly across both Samsung and TSMC-equipped iPhone 6ses. According to the sites, both devices are running the same version of iOS (9.0.2) and are 64GB iPhones.
One thing we want to note off the bat is that this doesn’t seem to be a performance problem.One article tries to argue that deviations in AnTuTu scores are signs that the Samsung devices aren’t as fast as the TSMC hardware, but the largest displayed gap between the two on any given run was just 2%. We don’t trust or recommend AnTuTu as a benchmark, but there’s still no sign that Samsung is losing performance to TSMC.
What there is evidence of, however, is that the Samsung device is using more power than its TSMC counterpart. The AnTuTu repetition finished 12 runs with the TSMC hardware at 77% and the Samsung product at 71%. The Javascript looping test showed the TSMC solution at 57% and Samsung at 42% after completing itself. That evaluation is somewhat more interesting, because the authors of that article didn’t just loop a website — they also ran an hour-long video playback benchmark.
Over the course of an hour of video, the Samsung-equipped iPhone went from 59% to 42% (a drop of 17%) while the TSMC device went from 74% to 57% — a drop of 17%. Video playback is going to be handled by fixed function blocks within the GPU, which means whatever power consumption issues are being seen are specific to CPU-heavy workloads. The question is, how much does this matter to real-world use cases?
Geekbench data seems to back this up: A Reddit thread suggests that Samsung devices are scoring just over six hours of battery runtime, compared to over eight hours from TSMC devices.
Evaluating the problem
So far, we know two things: Battery life on Samsung-equipped iPhone’s isn’t as good as the TSMC version when running looped benchmarks and the phone’s measured temperature when running those benchmarks is allowed to rise higher than the skin temperature of the TSMC device. I say “allowed” to rise because it’s likely that Apple monitors skin temperature to determine when the chip needs to throttle.
This situation could be deliberate, a result of Apple balancing poor characteristics of the Samsung solution to ensure equivalent performance at the cost of battery life and heat. It’s also possible that the Samsung devices need a firmware update to bring their thermals and battery life into line with its Taiwanese rival. It’s even possible that the increased heat and battery life were exacerbated by Samsung’s smaller die size. Samsung’s die-size advantage theoretically offers a better cost structure for Apple, since it can squeeze more dies out of each wafer, but closely-packed transistors tend to run at higher temperatures, all else being equal. Chips at higher temperatures draw more power, and may require more voltage to operate at the same frequency. These trends would negatively impact battery life.
This problem, however, could also explain why Apple second-sourced the chip in the first place. TSMC hasn’t hit the size targets that Samsung has, but they may have built the better SoC in this particular case. More investigation is needed, however, before we declare exactly what happened or what solutions exist, if any. It’s not clear how much this problem will impact devices in normal use cases — very few people spent an hour running JavaScript tests, and the video playback benchmark showed equal levels of power consumption between both products. It’s possible that this problem captures benchmark behavior rather than a huge consumer issue, though the fact that users went looking for answers suggests that skin temperature is noticeably different between Samsung and TSMC.
Update: In a statement to Ars Technica, Apple claims that the Geekbench battery tester is unrealistic because it spends far too much time in the highest-performing CPU state. “Our testing and customer data show the actual battery life of the iPhone 6s and iPhone 6s Plus, even taking into account variable component differences, vary within just 2-3% of each other.”
Those of you who argued that Apple’s response would amount to “You’re testing it wrong,” aren’t wrong. Then again, as we’ve recently covered, smartphone benchmarks have their own suite of problems.
Geek Blog 