Jonathan Gallmeier, AMD
-
Heterogeneous system architecture: Multicore image processing using a mix of CPU and GPU elements
Image processing is computationally intensive, requiring immense resources in CPU and memory throughput. Parallelism through multiple CPU cores helps, but even with today's dual-, quad-, and higher count core devices, multimedia tasks demand either a great deal of power or the addition of dedicated image processing hardware. Adding a pool of parallel processing elements as a programmable accelerator to the CPU creates a proper balance of general-purpose processing, high performance, and low power, as demonstrated in an example showing 2x-3x performance and performance-per-watt improvements.