Qualcomm Adreno 512 vs PowerVR GX6450

Qualcomm Adreno 512

The Qualcomm Adreno 512 is a mobile graphics card for mid-range smartphones and tablets (mostly Android based). It is included in the Qualcomm Snapdragon 660 SoCs and based on the Adreno 500 architecture (like the Adreno 520 in the S820, which should be fully compatible in software).

According to Qualcomm up to 30% faster than the previous generation (Adreno 510?).

The GPU supports modern standards like Vulkan 1.0 (according to Wikipedia), OpenGL ES 3.1 + AE (3.2 in other sources), OpenCL 2.0 and DirectX 12 (FL 12.1 according to Wikipedia, 11.1 according to Qualcomm). Furthermore, the GPU supports Universal Bandwidth Compression (UBWC) to save memory bandwidth.

PowerVR GX6450

The Imagination Technologies PowerVR GX6450 is a fast mobile graphics solution for smartphones and tablets, which can be found in ARM SoCs such as the Apple A8 (iPhone 6 and iPhone 6 Plus). The chip is available since Q3/2014 and of one of the fastest solutions in its class at the time of introduction.

Architecture and Performance

The GPU is based on the PowerVR Series 6XT (Rogue) Unified Shader architecture and integrates four clusters with two TMUs and 32 FP32-ALUs each. Therefore, the theoretical FP32 (full precision) compute performance per clock is identical to the previous G6430 (also 4 clusters) found in the Apple A7. Thanks to additional dedicated FP16-ALUs, however, the also important FP16 (half precision) performance has been greatly improved.

In the iPhone 6, the GPU should be clocked at around 450 MHz, delivering about 115 GLOPS (FP32). Compared to other mobile graphics solutions, the GX6450 outperforms the Qualcomm Adreno 330 (Snapdragon 800/801) and matches the more recent Adreno 420 (Snapdragon 805). Except for Nvidia's power hungry tablet chip Tegra K1 and its bigger cousin GX6650, the GX6450 is one of the fastest mobile GPUs in late 2014. Even the most advanced games and high resolution displays are handled without any problems. Among others, the GPU supports OpenGL 3.x/4.x, OpenGL ES 3.0 as well as DirectX 10.

Power Consumption

Thanks to the advanced 20-nanometer process and the relatively low clock rate, the GX6450 implementation in Apple's A8 offers a very high energy efficiency. On the other hand, the GPU has a comparatively "wide" architecture requirering a lot of transistors and die size (19 mm² in 20 nm).