News

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?