Technology to Help You Sleep Better at Night



Heterogeneous multicore designs help reduce system cost, size and power
consumption for real-time, high-performance systems worldwide.

While people may not realize it, they interact with or depend on high-performance multicore systems every day and indirectly impose their requirements. For example, in medical imaging these systems are required to receive the vast amounts of data from the sensors in MRI and CT scanners and then convert that data into an extremely precise image that can be analyzed by a radiologist. In video surveillance, many cameras capture images simultaneously, which must then be quickly correlated and analyzed to identify and track items or subjects. And in homeland security applications, vast sensor networks are deployed in ports to scan for hazardous or radioactive materials that might be smuggled into our country. Each of these applications needs to improve its real-time capabilities through increased performance. At the same time, there are desires to make these applications portable and more affordable – to improve healthcare by deploying medical imaging systems to field hospitals in impoverished countries; to improve security in locations where high-performance servers can’t easily be deployed for environmental or cost reasons; or to improve the nation’s security through better coverage of our coastal ports, both large and small.

Developers of these current-generation multicore applications have learned how to squeeze as much performance as they can out of their multicore platforms. But conventional sequential microprocessors and coding languages are reaching their limits for delivering the faster real-time performance, smaller system size and low-power demands of next-generation applications. Wider CPU registers, data paths, deeper caches, instruction-level parallelism, multicore processors, frequency scaling, ASICs and FPGA offload engines are all being stretched in their ability to deliver the required performance under increasingly stringent power budgets.

So developers are turning to new heterogeneous system architectures that are showing good promise at being able to deliver the performance they need while at the same time enabling reduced system costs, size and power consumption. These new architectures leverage traditional graphics processors, with their hundreds of parallel compute units, to offload certain tasks from the CPU. The graphics processor has evolved from a basic device to render an image on the screen, to the advanced 3D graphics processors that drive the rich visual interfaces of today’s PCs. Along with these advances in GPU capabilities have come advances in their programmability; over time, they have transitioned from driver-based programs to system-based programming models for graphics. Today’s GPUs are built around an advanced parallel processing architecture, which is capable of operating on hundreds of pieces of data in parallel in a single cycle. At the same time, open and royalty-free programming standards for general-purpose computations on heterogeneous systems, such as OpenCL, are going beyond just graphics processing to programming the GPU for system-level applications. Multiple GPU cards can already be ganged together to deliver combined graphics performance in PC applications. These same techniques are being utilized to build GPU clusters for high-performance computing (HPC) applications.

Developers of high-performance multicore systems are taking advantage of this new technology so that in the future we might all be able to sleep a little better at night, knowing that there is less to worry about from terrorists trying to smuggle weapons of mass destruction into our country, that Third World countries are getting better health care, and that our streets and facilities are more secure.

 


techno_help

Cameron Swen is a senior manager of product marketing in AMD’s Embedded Solutions Division. Swen joined AMD in 2003 as the manager of technical marketing for AMD’s Innovative Solutions Group. He started his career 18 years ago as an engineer working with embedded computer systems and has held a variety of technical marketing positions at National Semiconductor and AMD for the last 11 years. Swen holds a degree in engineering from Colorado State University.

Share and Enjoy:
  • Digg
  • Sphinn
  • del.icio.us
  • Facebook
  • Mixx
  • Google
  • TwitThis