The pace of innovation and performance increases in the PC graphics industry is tremendous. Instead of just keeping up with Moore’s Law, PC graphics double performance approximately every 18 months versus every 2 years (see Figure 1). At the same time, innovative features are added with each new generation. Although embedded systems do not change at nearly the same rate as the PC industry, embedded designers can take advantage of the fast-paced innovation in graphics technology to help design systems capable of productive use in the field for 5, 10, or even 20 years.
One of the most exciting new graphics technologies is the ability to drive up to six independent displays from a single Graphics Processing Unit (GPU). Found on the latest generation of GPUs, ATI Eyefinity technology can be used by a wide variety of applications to provide a large display surface with an area of up to 24 million pixels split across several monitors or independent content on each monitor, depending on the graphics card model. Simulation systems can use multiple monitors to offer peripheral vision and greater spatial awareness. Casino gaming systems can drive the main screen, entertainment screen, and player hospitality screen all from a single GPU. High-end digital signage systems can drive multiple screens without special-purpose display hardware.
Another display technology of interest to embedded developers is DisplayPort. The latest version of the standard, DisplayPort 1.2, can double the bandwidth of HDMI and support stereoscopic 3D. Embedded system developers can appreciate that DisplayPort is an open standard free of royalties. DisplayPort 1.2 also adds the ability to daisy-chain monitors with multiple independent video streams, thereby cleaning up the swarm of cables needed for multi-monitor systems such as display walls.
The graphics technology likely to have the greatest impact on embedded applications is General-Purpose computing on Graphics Processing Units (GPGPU), also known as stream computing. Today’s GPUs have evolved into massively parallel, programmable architectures optimized for performance per watt and per mm². By leveraging both the CPU and GPU for computation, developers have a broader toolbox to optimize and accelerate their applications. With the advent of the OpenCL programming standard for general-purpose computations on heterogeneous systems, software developers now have a way to use the same language for both CPUs and GPUs.
The parallel processing of stream computing is a great match to challenging problems in embedded applications with image and signal processing. The GPU can be used to accelerate encoding and decoding in video conferencing applications, creating the 3D model in medical imaging, the image formation process in radar imaging, pattern recognition in signal analysis, and seismic data processing for oil and gas exploration.
Embedded designers can access the first level of this technology with the Embedded ATI Radeon™ E4690 GPU (Figure 2). With 320 stream processors producing up to 3.88 GFLOPS of peak single-precision performance, the ATI Radeon E4690 GPU consumes only 25W and includes graphics memory in the same package.
The stream processing power is accessed through the ATI Stream SDK, including the compiler, device driver, performance libraries, and performance profiling tools.
The ATI Radeon E4690 GPU matches well with an AMD Turion™ processor to provide balanced serial and parallel computing as well as high-performance graphics and high-def video. The ATI Radeon E4690 GPU supports multiple types of display outputs, including DisplayPort, for flexible choices in embedded displays. For the next level of technology, one only has to look at the recently introduced ATI Radeon HD 5000 series of consumer graphics boards to see which technology may make its way into embedded graphics solutions.