One of the themes that has recurred in 2015 is that of System-on-Chip temperatures, primarily because of a story Qualcomm would like the media to forget - the high temperatures that their flagship processor, the Snapdragon 810, showed during both benchmarking and real-world use. There appears to have been a disconnect somewhere between the chassis designer not providing adequate cooling and the software engineers not considering the heat output of the chipset when running at maximum power, perhaps augmented by chipset heat specifications being optimistic from the designer. As the chipsets (and batteries) of devices heat up, so the software progressively underclocks the chipset - stopping the higher performance cores from running and reducing the maximum clock speed of the device. In effect, this meant that after running a moderate workload (such as browsing a number of webpages using the Chrome browser), devices based around some 2015 flagship processors were little quicker or responsive than those running a 2014 flagship processor.
For 2016, the chip designers are working on products designed to output less heat and smartphone designers have publically modified designs to improve cooling, but one interesting question is: how big an issue is this? How hot do our devices get when working hard? Testbird, a company interested in compatibility testing, have published some interesting statistics showing the arithmetic average (mean) System-on-Chip temperatures combined with the peak reported temperatures for an extensive list of tested chipsets. These tests are conducted using Testbird's normal procedure: the testing and monitoring applications are installed onto a stock device with the latest official software version, and tests are conducted using as many devices as is possible (up to three hundred) in order to obtain the most data. Most applications are tested for an hour, so in the case of games this includes opening the application, logging in, switching screens and similar. This ensures a degree of heat soak into devices, as whilst games rarely run a device at maximum performance all of the time, they work many components relatively hard.
The highest temperatures recorded in the table equate to the Qualcomm Snapdragon 200, at 63.9 degrees C, followed by the Marvell PXA1088 at 61.7 degrees. Other manufacturer chipsets only just exceed 50 degrees C. The Qualcomm Snapdragon 800 belongs in the list of chipsets with a maximum reported temperature over 50 degrees C together with the popular Snapdragon 400. MediaTek chipsets dominate at the lower end of the scale, with five chipsets populating the list, together with three Qualcomm chipsets, the HiSilicon Kirin 935 and the Samsung Exynos 5260. However, maximum recorded temperatures are just one part of the equation: the averaged reported temperatures are interesting too. Here, the top ten average temperatures are between 36 to 38 degrees C and the hottest chip is the Spreadtrum SC8810i, an older generation 1.0 GHz single core processor built on a 40nm process. Notably, the Qualcomm Snapdragon 805 appears in this list with an average temperature of 37.5 degrees C. At the bottom of the list, the average coolest running chipsets are split between four MediaTek and six Qualcomm processors, with the quad-core MediaTek MT6589 as being the coolest chip in the test group and having an average temperature of a shade over 26 degrees C.
Looking through the results, three chipsets stand out as having a wide difference between the average and peak temperatures: the Marvell PXA1088, the Qualcomm Snapdragon 200 and the Qualcomm Snapdragon MSM8528 (possibly belonging to the Snapdragon 400 series). This wide disparity might indicate that under certain load conditions, these chipsets run at a high temperature but are usually much cooler. Or it might indicate a compromised device design (with poor chassis cooling) and the processor is aggressively underclocked because of high temperatures. Unfortunately, we don't have information as to the applications or games being tested or the devices in use.
Another trend looking through the data is that most manufacturers show a similar average and peak temperatures, with Marvell and Spreadtrum showing a higher average and MediaTek and HiSilicon showing the lowest average temperature. Unfortunately, Testbird's information does not include the 2015 flagship chipsets, and Texas Instruments no longer manufacture chipsets. It would be interesting to see how the likes of the Exynos 7420 and Snapdragon 810, or the newer generation devices, cope with these tests. We also do not have full details as to what the tests are and if the workload is standardized or normalized - what screen resolution and detail were the games running in, as if a more modern device runs the application or game at a higher screen resolution, it is working the GPU harder, which could increase System-on-Chip temperatures. We also need to be aware that different devices will be better able to dissipate heat than other designs. Nevertheless, with an eye over historical data, the coolest chipset manufacturer is MediaTek.