Understanding Mobile Graphics Performance

By Laszlo Kishonti

Mobile graphics technology is developing rapidly behind the scenes. Choosing and implementing the appropriate graphics solution can make or break the success of new handsets.

When Kishonti Informatics started benchmarking mobile graphics at the end of 2002 it had only a couple of programmable mobile phones available, most of them supporting only the MIDP 1.0 Java Me specification with very limited graphics capabilities. Many of the LCD screens were black and white only with a resolution less than Quarter QVGA (120x160).

In order to help consumers and handset vendors evaluate the capabilities of tested devices, JBenchmark implemented a simple result upload function into our very simple first Java benchmark and encouraged all the users to send their results into our public web based database.

In those early days we were extremely enthusiastic about new possibilities in mobile computing and graphics, expecting hundreds of millions of people to be playing graphically rich games in the space of a few years.

Graphics technology – key element of product development

In the six and half years since launching the JBenchmark service, and with over two thousand devices benchmarked, we are now more enthusiastic than ever as mobile graphics has become even more important than we first thought. Whilst gaming is not now as important as we expected to be (possibly due to the screen size and physical dimensions of devices) graphics technology has found its place in a more fundamental way in handsets.

There is no serious handset or mobile chip vendor today who can think about graphics technology as a secondary factor when designing new products. Part of the solution is to embed hardware accelerated graphics (or GPU) into the devices. The reasons for this are the following:

1. Consumer expectations regarding the user interface design are far higher than two years ago. Users expect appealing and responsive applications and windowing systems which need efficient graphics solutions.
   
   
2. The screen resolutions of mobile devices are much more heterogeneous than in the desktop world. Designing applications and user interfaces for different and unforeseeable resolutions and aspect ratios is slow and painful. The problem can be solved using scalable 2D or 3D vector graphics which can be slow to process if using the device CPU only.
   
   
3. As people spend more time using mobile applications, battery efficiency becomes even more important. Hardware accelerated graphics is more power efficient than rendering the same content with the main CPU.
   
   
4. Certain applications like browsing, multi media and navigation are graphically intensive and can be efficiently accelerated by GPU-s.
   
   
5. As screen resolutions increase (partly because touch screen phones have more space for LCD screens) the number of color pixels to fill grows in a squared function. There is no chance to increase the main CPU clock frequency this way. The GPU has much higher fill rate and was designed to do this more efficiently.
   
   
6. Some simple and popular graphical effects make more efficient use of small screens but are prohibitively slow on CPU: such as alpha blending windows or images; perspective correction or warping textures; antialiased or multi sampled fonts and icons
   
   

Consistent with all these factors is that we see many new projects related to navigation and user interfaces and the trend is gathering momentum.

Cross platform benchmarking

The other major shift in our short history was caused by the emergence of the programmable smartphones. This made us enter the native benchmarking market and extend our support to all major mobile platforms - Brew, iPhone, Linux, OpenKode, Symbian, Windows Mobile, etc – and launching a new native performance database at www.glbenchmark.com in 2006.

As we have the same content running on all these platforms (and also on Java) our partners can make consistent cross-platform performance tests like comparing Android (running Dalvik Java) with iPhone (running OS X).

These cross platform benchmarks wouldn’t be possible if there were no industry standard graphics programming interfaces (APIs). Fortunately Khronos Group created OpenGL ES 1.x and 2.0 for 3D and OpenVG for 2D vector graphics which are generally supported by the industry. A similar process is taking place in the Java space through JCP creating SVG, M3G 1.0/2.0 and OpenGL ES binding APIs.

Supporting all these APIs consistently also makes possible to create cross-API benchmark tests, for example comparing OpenGL ES 1.x to OpenGL ES 2.0 or Java SVG to native OpenVG on the same device.

In the last six years, JBenchmark has adapted to the ever changing market environment through strong cooperation with our partners. Kishonti Informatics has developed a wider range of benchmarking tools to help its customers choose, test and debug graphics implementations and enter the market in time with stable, standards compliant and attractive solutions. Currently, only one percent of the mobile devices have efficient hardware accelerated graphics, we think this is only the beginning.