GPU And Gaming Performance
Mobile GPU performance is becoming increasingly important as people begin to see their phones and tablets as portable gaming machines. This section explores GPU performance with several synthetic and real-world game engine tests. To learn more about how these benchmarks work, what versions we use, or our testing methodology, please read our article about how we test mobile device GPU performance.
The PowerVR GX6450 GPU in the iPhone 6 Plus is the direct descendent of the G6430 GPU used in the ZenFone 2 (also used in the previous generation iPhone 5s). These GPUs have the same ALU performance, processing 256 FP32 FLOPs/clock. The ZenFone 2 runs its GPU at either 457MHz or 533MHz under load, and while we do not know the exact peak GPU frequency for the iPhone, it does lie within this range. This means we should see similar performance in pixel shader bound workloads. Where the GX6450 gains an advantage, is in front of and behind the unified shader cores, improving vertex processing and pixel fill rate, respectively.
In 3DMark, the ZenFone 2 keeps pace with the flagship phones. The iPhone 6 Plus leads the pack in the first graphics test that focuses on vertex processing, clearly benefitting from its front-end architectural tweaks. The ZenFone 2 is 19% slower but only 12% slower than the speedy Galaxy S6.
Qualcomm’s Adreno GPUs perform well in ALU-bound workloads, so it’s no surprise to see the Adreno 418 in the LG G4 post the highest frame rate in the second graphics test that focuses on pixel processing. The ZenFone 2 is 18% slower than the G4 and 15% slower than the iPhone 6 Plus.
The 3DMark Physics test focuses on CPU performance and uses a data structure that requires random memory accesses. As we touched on in the AndEBench memory latency test, the memory controllers in the iPhone’s A8 SoC and G4’s Snapdragon 808 SoC are optimized for serial data access patterns, which penalizes them heavily in this test despite their fast CPUs. Intel’s Atom SoC embraces random memory patterns, propelling the ZenFone 2 to the front of the pack, beating even the S6. This is one of the reasons it performs so well with real-world apps.
Basemark X is a very demanding benchmark, heavily stressing the entire graphics pipeline, especially vertex processing. At the medium quality setting, the ZenFone 2 is only 21% slower than the LG G4, but it cannot keep up with the other flagship devices, coming in 40% slower than the iPhone 6 Plus. It does manage to outperform the G4 in the onscreen test, where the latter device gets bogged down by its QHD resolution.
The order does not change at the high quality setting, although the gap between the ZenFone 2 and the flagship phones closes a bit; it’s only 11% behind the G4 and 25% behind the iPhone 6 Plus. Fed by bountiful memory bandwidth, the Mali-T760MP8 in the Galaxy S6 is a vertex processing machine, propelling it to the top of the chart.
The OpenGL ES 3.0 based Manhattan test makes heavy use of pixel shaders and other post-processing effects. As we’ve seen in the previous tests, the ZenFone 2 has no trouble outperforming the Adreno 306 and 405 found in the Moto G (3rd gen) and Xperia M4 Aqua. Although it’s not shown in our chart, the ZenFone 2 also performs better than Snapdragon 801 powered devices such as the Galaxy S5 and OnePlus One using the Adreno 330 GPU. For reference, the ZenFone 2 is 30% slower than the iPhone 6 Plus when running Manhattan.
T-Rex results mirror those from the Manhattan test, including the ~30% deficit to the iPhone 6 Plus.
We mentioned previously that the PowerVR G6430 GPU in the ZenFone 2 and the GX6450 GPU in the iPhone 6 Plus should have the same ALU performance, an observation confirmed by the GFXBench 3 ALU test. It also comes very close to matching the ALU performance of the Galaxy S6. Currently, the Adreno GPUs have the advantage in FLOPs/clock, which is why the LG G4 tops the chart in this test.
The Mali-T760MP8 GPU in the Galaxy S6, along with the two PowerVR GPUs, can all process up to eight texels per clock. However, the ZenFone 2 falls behind the iPhone 6 Plus in the Fill test by 14%. It’s likely that the iPhone’s GX6450 has larger texture caches as one of its back-end improvements. In addition to some architectural differences, the Mali GPU in the S6 runs at 700MHz versus the ZenFone 2’s 533MHz, increasing its advantage.
With its last generation GPU, the ZenFone 2 falls behind the latest flagship devices. Compared to the iPhone 6 Plus, the Asus phone is roughly 30% slower (15% to 40% based on our tests). However, it does perform better than other phones in the same price range, in some cases considerably better. Even more impressive, the less expensive version of the ZenFone 2 that comes with the Atom Z3560 SoC uses the same PowerVR G6430 GPU tested here, giving you gaming performance similar to the iPhone 5s for only $200.
Throughout this section, we kept comparing the ZenFone 2 to more expensive, flagship devices. This may seem unfair, but it’s actually a compliment. The midrange ZenFone 2 just performs more like a low-end flagship than the lower-cost phone that it is.