Battery Life

Battery life remains one of the most important aspects of any mobile device. After all, you can’t really call something mobile if it has to spend most of its time connected to an AC adapter. As a result, battery life testing is one of the most critical aspects of our testing, and it’s something that we spend quite a lot of time discussing internally.

Before diving into our results, I want to start things off talking about testing methodologies. This year we're implementing an overhaul of our web browsing test for battery life, with the Galaxy S7 review being our first chance to deploy it. As far as our long-standing 2013 test goes, at a high level our 2013 test was relatively simple in the sense that it simply loaded a web page, then started a timer to wait a certain period of time before loading the next page. And after nearly 3 years it was time for it to evolve.

Internally, we’ve been discussing reasonable measures to push our web browsing test in new directions to both better represent real-world workloads in addition to ensuring that we’re testing more than just display power. For at least a few devices, it had already become quite evident that our old test was almost purely display-bound to such an extent that even video playback was more power intensive. Other issues that were raised both internally and externally included the fact that the test would not test aspects like CPU governor boosts upon touching the display, and that our test almost entirely ignored things like 2D drawing and display pipeline efficiency.

In recognition of these issues, we’ve spent the past few months working on a new test. In addition to new webpages that are exact copies of many popular websites today to better represent modern, real-world workloads, we’ve added a major scrolling component to this battery life test. The use of scrolling should add an extra element of GPU compositing, in addition to providing a CPU workload that is not purely race to sleep from a UI perspective. Unfortunately, while it would be quite interesting to also test the power impact of touch-based CPU boost, due to issues with reproducing such a test reliably we’ve elected to avoid doing such things.

However, we don’t take these changes lightly. While we’ve validated the workload for several devices, it’s important to emphasize that these results could change in the future as much of this data is preliminary. For the second part of the review I’ll be sure to revisit these results with an expanded dataset. Of course, other than the workload the device setup has been held constant across these tests by equalizing brightness to 200 nits and disabling all background sync to the best of my ability.

Web Browsing Battery Battery Life 2016 (WiFi)

As we can see in the results, the Galaxy S7 and S7 edge both do impressively well. One of the more interesting comparison points here would be against the latest devices like the Huawei Mate 8, which has the Kirin 950.

Our previous test was relatively display-bound so differences in SoC efficiency were often difficult to discern and often masked entirely, but here we can see an enormous spread that is almost entirely due to SoC efficiency. The Huawei Mate 8, which under our previous test seemed to be only slightly above the iPhone 6s Plus, has gained a noticeable lead in this test as the Kirin 950’s CPU efficiency is ahead the competition at this time, although it’s important to keep in mind that CPU efficiency is not the only relevant metric for an SoC.

Interestingly enough, the Galaxy S7’s battery life is almost directly scaling with battery size relative to the Galaxy S6. As we’ll see in the display section, the Galaxy S7’s display is pretty much identical to the Galaxy Note5 and S6, so it looks like the efficiency gains from the Galaxy S6 to the S7 are small if you look at the Snapdragon 820 variant.

Of course, the big question that I’m sure a lot of people are thinking is how the Galaxy S7 and the Snapdragon 820 compare to the iPhone 6s and 6s Plus. Unfortunately, due to timing constraints we weren’t able to get data for the smaller iPhone 6s, but looking at the iPhone 6s Plus relative to the Galaxy S7 edge it’s pretty obvious that there is a power efficiency gap between the two in this test. Despite the enormous difference in battery size - the Galaxy S7 edge has a battery that is 33% larger than the iPhone 6s Plus - the difference in battery life between the iPhone and Galaxy in this test is small, on the order of half an hour or 5-6%. This is balanced against a higher resolution (but AMOLED) display, which means we're looking at SoC efficiency compounded with a difference in display power.

Web Browsing Battery Life (WiFi)

In the interest of providing another data point and some validation of our testing results, I ran both devices through our old web browsing test to see what the results would be for something that should be display-bound. Here, it’s obvious that the Galaxy S7 edge holds a significant lead over the iPhone 6s Plus, although the use of a higher resolution display and an AMOLED display in a high-APL test means that the GS7e is using more power in this test as well. However, when you take these results with our new web browsing test, it becomes evident that a difference in power efficiency is growing as the load on the SoC grows. Similarly, despite the Galaxy S7 being neck and neck with the Huawei Mate 8 in our older test, it loses the lead in our new web browsing test.

Of course, I have caution that all of the data that we’re gathering for the web browsing test is still subject to change, but given the interesting data that it provides it’s important for us to include these results, as we’re reasonably confident that these results are accurate.

Overall though, it’s clear that the Galaxy S7 and S7 edge will have solid battery life, even if device efficiency isn’t quite on par with the very best that we’ve seen so far. An improvement of 15% is going to be noticeable if you upgrade from the Galaxy S6, and anyone upgrading from a phone with an SoC older than the Snapdragon 800/801 generation will see huge improvements here.

Introduction and Design SoC and NAND Performance
POST A COMMENT

202 Comments

View All Comments

  • TheinsanegamerN - Wednesday, March 09, 2016 - link

    It will let us know if throttling is a big issue. Benchmarks are just that, tests to see how well the phone does.

    Of course we dont run benches all day, but if you do not run benches, then how do you know how well the phone does when pushed, in say, a mobile game like asphalt 8 for more than 5 minutes? It's useful information to have.
    Reply
  • tuxRoller - Saturday, March 12, 2016 - link

    Well then, according to that measure, throttling is a big issue with the iPhone 6s.
    How do you tell how well it works with asphalt 8? I'd probably pay asphalt 8 for awhile.
    These benchmarks aren't a good proxy as they are designed to push the devices to their limits.
    Reply
  • tuxRoller - Tuesday, March 08, 2016 - link

    Also, XDA ran some graphics tests continuously if you're interested.
    All of the phones throttled, save the note 5.
    Reply
  • frenchy_2001 - Tuesday, March 08, 2016 - link

    here is the xda article:
    http://www.xda-developers.com/s7-edge-throttling-t...

    Note that the note5 did not throttle in GPU test, but its final run was still only 60% of the throttled S7 score (and about 50% of the starting score).
    Reply
  • tuxRoller - Tuesday, March 08, 2016 - link

    I just mentioned that it didn't throttle since jjj seems to be an exynos fan, so I was trying to forestall "BUT THE EXYNOS DIDN'T THROTTLE", even though it had lower results throughout. Reply
  • gijames1225 - Tuesday, March 08, 2016 - link

    Great write up. Glad to see the Snapdragon 820 is properly flagship level. I look forward to the rest of the review, details on throttling, and hopefully at some point a look at Samsungs new Exynos SoC in the other s7 model. Reply
  • Drumsticks - Tuesday, March 08, 2016 - link

    It's kind of interesting that Apple has a solid win in CU performance against Qualcomm, but a loss in GPU, even though they get their GPU from a company who should specialize in them. Twister truly is an astounding architecture, but they're probably at the end of the easy, huge performance boosts. It may be interesting to see how well Apple can go about completely redesigning their core when it happens. Reply
  • name99 - Tuesday, March 08, 2016 - link

    Faster GPU is not as interesting as faster CPU because it's trivial to make a faster GPU --- just throw more GPU "cores" at the problem. If Apple wanted to double the GPU performance of the A9 tomorrow, all they have to do is use 12 GT7000 cores instead of their current 6.
    So the decision as to how many to use is an economic/use case decision, not a technical one. Given the pixels Apple is interested in driving, it looks like they used the right number of cores. QC sells into a broader market (and, in particular, a market that, whether or not it makes sense, uses crazy-high pixel densities) so their incentives align with throwing in more cores.

    If you want to run a technically interesting GPU performance competition between Apple and Adreno (or Mali, or anyone else), raw performance is not the interesting metric. The interesting metrics are
    - performance/watt
    - performance/mm^2
    I don't know who would win either of these, but my guess is Apple. We don't yet have enough information to answer the first question, but the mutterings about GPU throttling in the comments suggest that QC gets hig GPU numbers by burning lotsa power, not by having some super-advanced GPU that's a cleverer design than anyone else. In a sense, what you're seeing is what Apple could copy if they wanted by putting the A9X inside the iPhone6S.
    Reply
  • lilmoe - Tuesday, March 08, 2016 - link

    " The interesting metrics are
    - performance/watt
    - performance/mm^2"

    Nah, the really interesting metric would be how much power consumes to run (and sustain) the average game at 60FPS at the same resolution. Read: efficiency.
    Reply
  • sor - Tuesday, March 08, 2016 - link

    Efficiency... So performance/watt. Reply

Log in

Don't have an account? Sign up now