Chinese chipset vendor Rockchip has just joined the ‘high efficiency, low power’ club by recently announcing the 28nm RK3168 dual core processor with PowerVR SGX graphics for mobile devices.
The RK3168 platform was displayed at the 2013 spring edition of the Hong Kong Electronics Fair, where Rockchip had a tablet reference design connected to a power supply which showed the overall power consumption. The device was running at an average of 280mA @ 3.8V (about 1 W) while playing a 1080p video, with the display brightness set to about 50%.
The Rockchip RK3168-based reference platform playing back a 1080p video
Rockchip RK3168 – another chapter in the PowerVR SGX success story
The PowerVR SGX540 GPU inside Rockchip’s RK3168 platform might be familiar to many from its other incarnations inside other popular chips such as the Texas Instruments OMAP4430 (recently included inside Google Glass) and OMAP4460, Ingenic JZ4780, Samsung Exynos 3 Single, Intel ‘Medfield’ Z2460 and ‘Lexington’ Z2420, and many others.
The PowerVR SGX family’s efficiency as well as its ability to be successfully integrated at a number of process nodes and within a range of target frequencies has helped it build a reputation as the most successful OpenGL ES 2.0 solution available in the market. As recent measurements from independent reviewers have shown, PowerVR SGX GPUs are able to deliver the same high performance across a range of operating systems without requiring complex power management schemes that limit their throughput by having to severely down-clock the whole system when the SoC overheats.
In fact, heat measurements have proven that SoCs integrating our PowerVR graphics IP stay cooler throughout various use cases, including situations where the GPU is heavily used such as 3D applications, and therefore do not need regular interventions from power or frequency controllers.
Superior image quality with PowerVR GPUs
PowerVR SGX GPUs have FP32 ALUs, providing high precision processing throughout the graphics pipeline.
You might be surprised to find out that FP32 support is not as widespread as you think for other GPU architectures. In fact, there are currently at least two competing multicore solutions that dazzle customers with high core counts but still use less than FP32 precision in their ALUs, resulting in the lowest image quality in the mobile industry.
YOUi Labs has recently released an Android app which tests the floating point accuracy of the GPU inside your device
To give you an idea of how differences between FP32 and lower precision ALUs can impact gaming, we ran the Relative Benchmark on some tablets and smartphones around the office. This Android graphics benchmark runs an animation filled with post processing effects like HDR (High Dynamic Range), bloom, and depth of field that test mobile GPUs to their limit.
Visual differences between PowerVR SGX (FP32) and competing multicore GPUs (FP16) due to computational precision
On top of that, the PowerVR on-chip colour buffer is always worked on at full internal precision (RGBA8888), regardless of the final frame buffer precision. That means blending happens at high precision, reducing banding, while dithering is only applied once when the colour buffer is flushed to the frame buffer.
As you can see, some competing GPUs use 16-bit (RGB565) formats for internal colour computation which can cause a noticeable loss of precision after multiple reads and writes when blending (be it from external memory or on-chip tile memory). This may result in more significant banding and dithering artefacts.
The difference is that PowerVR’s tile buffer is always RGBA8888, irrespective of what the app asks therefore we avoid the ‘grainy’ look below as the dither pattern accumulates.
A scalable roadmap from mass market to high end
Because our PowerVR unified shader architecture was designed with GPU compute in mind, Imagination offers OpenCL support across all their PowerVR GPU cores. Integrating the PowerVR SGX540 GPU in their chip has made Rockchip’s RK3168 the only platform with GPU compute support across their RK3xxx range of SoCs.
This brings a very important advantage for consumers, especially in emerging markets. As our GPUs are the only processors in the market which are capable of handling both graphics and compute APIs across their entire range, platforms integrating Imagination’s solutions can still offer high-performance graphics and compute even on mass market devices, something unheard of or even impossible in competing graphics solutions.
Furthermore, Rockchip are benefiting not only from the most efficient graphics and compute hardware IP engine on the market, but also tapping into an ecosystem of over 35,000 developers with whom Imagination are always collaborating on application bring-up and optimization.