Skip to main content

ARM Vs. x86: The Secret Behind Intel Atom's Efficiency

Connecting The Dots With Web Browsing Power Use

Microsoft Surface
Load Time (ms)Platform (W)CPU (W)GPU (W)Memory (W)Panel Backlight (W)Everything Else (W)
Standard Page (MSN Homepage)9413.390.090.270.351.081.60
Standard Page + Split Screen13984.430.540.540.451.121.79
Complex Page (Long Article on Wikipedia)17703.460.100.270.321.111.63
Complex Page + Split Screen20164.240.360.430.511.101.83
Acer W510
Load Time (ms)Platform (W)CPU (W)GPU (W)Memory (W)Panel Backlight (W)Everything Else (W)
Standard Page (MSN Homepage)5802.940.070.120.311.181.26
Standard Page + Split Screen7763.
Complex Page (Long Article on Wikipedia)12593.
Complex Page + Split Screen14073.370.220.150.351.141.51

First things first. Although these uber-granular numbers were generated in Intel's lab, we can at least confirm that, using Chrome, Safari, or IE10 on Samsung's ATIV Smart PC 500T, the performance of Atom on Windows 8 is greatly superior to Tegra 3 on Windows RT. We haven't published our review on Samsung's Atom-powered tablet yet, but check out the following chart:

Now, with that in mind, the Atom-based Acer also delivers better power performance. Moreover, it's clear that the complexity of a page being displayed affects consumption, Intel's Atom demonstrating its strengths in the GPU and memory categories as Nvidia's ARM-based cores use slightly less power on the more complex webpage. 

This difference can also be seen in the raw results from the test equipment.

Google's homepage: Note the lack of activity

MSN's homepage is a lot busier.

The first screen shot shows the power consumption of Google’s home page and the second illustrates the power consumption of Seeing these graphs and their corresponding power levels should make it clear how difficult it is to assign just one number to a tablet's power use when the actual draw fluctuates pretty wildly. In the actual split-second that the screen shot of MSN's home page was captured, Acer's W510 happened to be using 2.56 W and the Surface was using 4.29 W. The average of both gets calculated to be 2.94 W and 3.39 W (which is how we're reporting it in the table).

It's really interesting that these results come close to matching each manufacturer's advertised specifications. Acer says its W510 will do 9 hours of battery life. If you average the 2.94 W and 3.28 W of power consumption for the first two test cases (3.11 W), you get 8.6 hours. Microsoft says its Surface will do 8 hours of battery life. And if you average the first two rows of the Windows RT-based tablet, you get an average of 3.91 W; a 31.5 Wh battery should deliver 8 hours.

Now, in our own calibrated 200-nit test (which adds MP3 playback in the background), the Surface achieves an 8:03 run time. Asus' Transformer Prime pulls off 8:01 (the same as the Surface), but with a 25 Wh battery instead of the Surface’s 31.5 Wh power source. So, the Transformer Prime’s consumption is probably around 3.12 W. As we saw on the previous page, compared to the Surface, we suspect the Transformer Prime is thriftier with power than Microsoft's tablet, which we'd again be inclined to attribute to Android versus Windows RT and each platform's display.

Time-Out For Some Speculation

Now, let’s speculate a little bit. Intel's decision to use a somewhat-complex webpage in full- and split-screen modes seems to match the way tablet vendors are rating battery life. Call it an unofficial "we don't want to get sued for false advertising" standard.

Andrew’s own numbers generated during our Surface testing (specifically, light Web browsing with Wi-Fi enabled and an MP3 playing back at 200 nits) suggest we'll see 8 hours 3 minutes of run time. In other words, our workload comes pretty close to what vendors seem to be using to test internally. Our test is probably a little more taxing, since we get 9 hours from our iPads instead of the advertised 10, 10 hours from a Kindle Fire HD instead of the advertised 11, and 9 hours from the Nexus 7 instead of the advertised 10.

On the iPad 2, we get 9.08 hours from a 25 Wh battery (solving for W gives us 2.75 W). For the third-gen iPad, 9.25 hours from a 42.5 Wh battery translates to 4.59 W of average power consumption. From this, we learn that the iPad 2 offered superb efficiency that wasn't matched by Apple's successor. We haven't run these numbers on the fourth-gen iPad yet, but power consumption should be similar as the iPad 4 also has a 42.5 Wh battery rated for 10 hours of mixed use.

Testing the power consumption of MSN's home page.

Google claims claims 7 hours of Web browsing on its Nexus 10 from a 33.75 Wh battery (so, 4.82 W for a 4 MP display) and Samsung's Series 3 XE303C12 Chromebook claims 6.5 hours from a 30 Wh battery (4.62 W for a 13x7 display). If this means paying an extra 0.20 W for a 4 MP display, and we go back to our previous page speculating that 3.75 W is what the Nexus 10 consumes during video playback, then the Chromebook probably will need 3.55 W for video playback running on a standard-resolution display. This also shows us that the fourth-gen iPad with the Swift-powered A6X is more efficient than A15, per our back-of-the-envelope calculations.

I’m willing to bet that the Cortex-A15, in low-power applications like video, is on par with the Atom. But in higher-power applications (ironically, Web browsing), it'll require 50% or so more power than the Atom. We know this must be true to some degree, since Google actually tells us in its advertising that you get less battery life from browsing than watching video. These numbers also suggest that Cortex-A9 is more efficient than -A15 under certain conditions, which is to say that even though -A15 improves performance, we need to see how much real-world work is done within a set thermal threshold.

Just so that Qualcomm doesn’t feel left out, we can probably speculate about its Snapdragon S4, since we know it is being advertised with a vague "up to 10 hours" of run time with a 28 Wh battery on Dell's XPS 10. That’s a best-case 3.11 W, which is what Acer's W510 requires when we average those first two Intel workload numbers.

At the end of the day, Intel's numbers make sense. It's pretty clear that the Atom is on par with today's top SoCs. At least under Windows 8, it's able to turn in better efficiency results than Tegra 3 under Windows RT, and should come close to Tegra 3 under Android, in light of our Transformer Prime numbers. 

When it comes to Web browsing, our best predictions are:

  1. Atom is neck and neck with the power consumption of Qualcomm’s Krait architecture in the Snapdragon S4.
  2. Atom beats Cortex-A15 on power consumption.