Results: GFXBench 3.0
Kishonti GFXBench 3.0 is a cross-platform GPU benchmark supporting both the OpenGL ES 2.0 and OpenGL ES 3.0 APIs. It comprises both “high-level” game-like scenarios, along with more “low-level” tests designed to measure specific subsystems.
The OpenGL ES 2.0-level T-Rex HD, a holdover from GFXBench v2.7, simulates a modern game using several complex effects, including motion blur, parallax mapping, and complex particle systems.
The low-level tests include Fill, which measures fill rate by rendering four layers of compressed textures; Alpha Blending, a test that renders layers of semi-transparent quads using high-resolution, uncompressed textures; ALU, for measuring shader compute performance; and Driver Overhead, which measures the CPU overhead of the graphics driver and API by making a lot of draw calls and state changes. See GFXBench 3.0: A Fresh Look At Mobile Benchmarking for a complete test-by-test breakdown of this benchmark.
In the synthetic tests we’ve seen so far, the Adreno 330 and Rogue GPUs offer similar performance. However, in the T-Rex benchmark, which attempts to demonstrate the performance of a typical, modern game, the iPhone’s Rogue crushes the Adreno 330 in the Note 3 and Nexus 5. Its score, no doubt, is boosted by rendering the fewest number of pixels.
The N1 manages to finish ahead of both the Mi3 and MX3, but is 2.5 times slower than the iPhone 5s and 1.6 times slower than the Note 3. In this real-world gaming test, there’s no performance difference between ColorOS and CyanogenMod.
Taking the T-Rex rendering off-screen doesn’t change the order at the top of the chart, but does significantly close the gap to the iPhone 5s. With everyone processing the same number of pixels, we again see very similar performance between Adreno 330 and Rogue.
Even though the MX3 and Mi3 render the same number of pixels on- or off-screen, they both see a frame rate increase when rendering off-screen, and even swap positions in the rankings. It appears Samsung’s Exynos 5 Octa and Nvidia’s Tegra 4 encounter a bottleneck between the frame buffer and display.
Rogue’s poor fill rate performance holds the iPhone 5s back in this test, allowing the Note 3 and Nexus 5 to claim the top spots. Even the slower Adreno 320 in the N1 is competitive with Rogue. For comparison, the N1 is about 1.5 times slower than the Note 3. The performance differential between the Adreno 320 and 330 seems to be hovering around 1.5x.
Running the Fill test off-screen produces results virtually identical to the on-screen benchmark.
Once again, Rogue’s poor fill rate performance holds the iPhone 5s back. The results in this metric essentially mirror those in the Fill test, which makes sense due to the functional overlap between them. Even the 1.5x performance delta between the N1 and Note 3 remains.
The only real difference is the Mi3, where the 1.3x deficit between it and the MX3 in the Fill test grows to 2.4x in Alpha Blending. It’s possible the Mi3 is using slower RAM. While I was unable to determine the speed rating of the SK Hynix module, the Geekbench 3 Memory test helps corroborate my theory. In the Single-Core test, the performance delta between the Mi3 and MX3 is again 1.3x, and in the Multi-Core test, where memory bandwidth is further taxed, the delta grows to 1.5x. The Alpha Blending test requires even more bandwidth, so it makes sense that the Mi3 performs worse.
Again, there’s no real change when running this test off-screen.
The on-screen ALU test is v-sync limited; the iPhone 5s, Nexus 5, and Note 3 all max out. The N1 comes close to the ceiling, but the Mi3 and MX3 fall significantly behind.
Rendering off-screen removes the v-sync constraint and yields more interesting results. The Adreno 330 shows off its shading prowess, easily defeating Rogue. The Note 3 is 1.17 times faster than the Nexus 5 and 1.79 times faster than the iPhone 5s. The N1, Mi3, and MX3 all fail to improve their scores.
The Driver Overhead test shows the iPhone 5s with nearly twice the performance, or almost half of the overhead, of the Mi3 in second place.
The N1 registers 21 percent less overhead when running ColorOS than CyanogenMod. Perhaps Oppo includes some driver tweaks to help performance. If so, they don't produce any tangible benefits in graphics performance.
The off-screen results are nearly identical, except the iPhone 5s manages to improve its already-impressive score.