CPU Performance Gains
Here's where things start to get interesting. I already established that general operating system performance feels slightly better in Android 5.0, so it's now time to quantify any performance improvements. Google says there should be a noticable performance improvement from moving to ART, even in benchmarks, and even with standard 32-bit instructions. How noticeable these improvements are in practice remains to be seen.
On this page we're looking at improvements in CPU and system bound benchmarks. As a refresher, the Moto X 2014, Galaxy S5 and LG G3 come with Snapdragon 801 SoCs inside with 2.5 GHz quad-core Krait 400 CPUs and Adreno 330 graphics. The LG G3 is the only device with 3 GB of RAM and a 1440p display, with the Moto X and Galaxy S5 coming with 2 GB and 1080p displays.
The Moto G 2014 is a mid-range device with a Snapdragon 400 SoC inside, with its quad-core Cortex A7 CPU clocked at 1.2 GHz. It comes with 1 GB of RAM and an Adreno 305 GPU to power a 720p display.
Note here that the LG G3 did not complete Basemark II 2.0 in Android 4.4 correctly. It did complete correctly in Android 5.0.
The Samsung Galaxy S5 and Moto G 2014 gained performance from their upgrade to Android 5.0, recording improvements of 5% and 6%, respectively. For sustained, high performance workloads without any change in internal hardware, this is a decent enough gain, though far from groundbreaking. The Moto G stands to gain the most here due to its limited hardware.
The LG G3 reduced in performance by 4% on average. I suspect this has to do with LG tweaking the CPU governor to be less aggressive, as with Android 4.4 the device did get very hot during benchmark runs, and was often thermally throttled. By allowing the CPU to run at slightly lower speeds, sustained performance can improve. And this was the case, with the G3 performing consistently when benchmarks were performed back-to-back.
The Moto X 2014 did not gain performance at all. It already performed very well in Android 4.4, beating most other flagships with similar hardware by a small margin. While some scores dropped after the upgrade to Android 5.0, strong performance in Vellamo and especially PCMark kept it alive.
Interestingly, PCMark for Android was the only benchmark which saw performance gains on every handset after switching to Android 5.0. Gains were particularly large on the Moto G at 18%, and I suspect this is due to PCMark's use of simulated real-world tasks. As I mentioned earlier, general OS performance seems smoother in Android 5.0, and PCMark might be illustrating this.