[Image Processing Engineer Column] Misunderstandings about Speeding Up
アポロ精工 アポロ精工 新横浜オフィス（旧ケーアイテクノロジー/旧市川ソフトラボラトリー）
We develop and provide solutions through software and hardware, but it is not uncommon for misunderstandings to arise from our guidance on "acceleration using FPGA." The misconception is that any processing implemented in software can be "accelerated" by developing a dedicated board. In reality, the types of processing that can be accelerated using FPGA are limited. It is suitable for processes that can execute the same calculations in large quantities simultaneously. An example of this is convolution filters. Even if you perform addition and multiplication operations simultaneously, by equipping the necessary number of multiplication and addition circuits, you can obtain results in one clock cycle, regardless of the range. Additionally, for processes that allow "pipelining," in software, you would perform process A, then process B, and then return to process A. However, in hardware, when transitioning from process A to process B, you can perform process A simultaneously, which reduces the processing time of the faster operation. Even with a 5-stage pipeline, results can be obtained in the time of just one stage. By the way, when accessing large volumes of image data, GPUs with very wide data bus widths are also effective. *News is distributed through our company newsletter.
Related product information
