Snapdragon 810 and Helio X20 in Inconclusive Thermal Tests

May 19, 2015 - Written By David Steele

The Qualcomm Snapdragon 810’s heat woes have been well documented over the year to date. The crux of the matter is that Qualcomm’s current flagship processor requires more cooling than the devices using it have been able to provide for it to deliver the site of performance that it should be capable of. The smartphones running the 810 are only a little quicker than predecessors and noticeably warmer in the hands in use. It’s not that the Qualcomm flagship is a slow processor, it’s that the processor has to underclock to avoid becoming too hot. Today’s story alleges to show a comparison between MediaTek’s new Helio X20 System-on-Chip and the Snapdragon 810, running on an Android 4.4 Kit Kat test device that lacked cellular and Bluetooth connectivity, using an 8 Mbps WiFi connection. Unfortunately, the information about the testing is to not sufficient to draw a solid conclusion but I will write about this later in the article.

First, let me introduce these two System-on-Chips. The Snapdragon 810 is a conventional big.LITTLE based around a quad core 1.5 GHz ARM Cortex-A53 processor paired up with a 2.0 GHz ARM Cortex-A57 quad core set of cores. The A53 core is the low power, low heat, high efficiency core and the A57 is the higher performance core for the heavier processor loads. The Snapdragon 810 will switch between the two clusters depending on workload and temperature. The X20, however, is a big.LITTLE design consisting of three clusters: there’s a quad core 1.4 GHz Cortex-A53, a quad core 2.0 GHz Cortex-A53 and a dual core 2.5 GHz Cortex-A72 core. I’ll discuss the configuration of these two SoCs later in the article.

Unfortunately, the tester presented the temperature data on a chart with unequal scales, which highlights that this is not an official test. The testing methodology is said to have included ten minutes of “casual Wi-Fi browsing,” ten minutes of Asphalt 8 (using the top quality graphics available) and ten minutes of Modern Combat 5. The test appears to have been run once on each test device. After the first test, both SoCs were springy at the they degrees point. By the second stage, the Snapdragon had reached thirty eight degrees compared with thirty three degrees for the MediaTek and after the third stage, the MediaTek’s temperature teens remained at this point compared with forty five degrees for the Snapdragon 810. However, whilst these temperatures are academically interesting it’s the claims about how the processors were managing their cores that is not interesting.

The source website states that during the WiFi browsing stage, the Snapdragon 810 used the A53 cluster and the X20 switched between the 1.4 GHz and 2.0 GHz A53 clusters. During the second stage, the Snapdragon 810 switched up to the A57 cores in order to benefit from their increased performance. The Helio X20 relied on the 2.0 GHz A53 cluster. During the third stage, the Snapdragon continued to rely on the A57 cores whereas the Helio switched up to the A72 core, claimed because of thermal management (in other words, to keep the SoC cool). And here’s where things don’t quite add up: the Cortex-A72 is a powerful processor that’s most efficient, from a power / performance perspective, when it is being worked hard. This produces plenty of heat. Under a similar load to the A53, the source website reasons that the A72 will run cooler and it might if the processor is able to race to idle. By race to idle, we mean that the processor completes the task as quickly as possible and then falls asleep until it needs to do something else. However, if the A72 is simply being used to keep the overall chipset running cool, this seems to make the idea of the three clusters somewhat redundant – and adds a layer of complexity into the SoC. It is just as likely that Modern Combat 5 predominantly used the 2.0 GHz A53 cluster and perhaps occasionally relied on the A72 cores, which is why the temperature remained at thirty three degrees.

At this juncture, we simply don’t have enough relevant information. Screen resolution, memory, detail settings for both devices, frames per second would all add to this test. Even a nod towards the browser and websites visited. And I have to ask why the test was conducted in 32-bit Android 4.4 when both chipsets are 64-bit compatible and surely a manufacturer would not dare release a device based around one of these chips running Android 4.4? We will report back when we have a better comparison.