Showing posts with label visualization. Show all posts
Showing posts with label visualization. Show all posts

Sunday, December 8, 2013

Memory Access Visualization

It is difficult for the human mind to comprehend large differences in magnitude, and we often forget what large magnitudes exist within computers.  Even creating graphics to effectively display these differences can be challenging.  I've generated some visualizations of memory access patterns of a test program running on my GPGPU design, which renders a cube with bilinear filtered textures applied to the sides.  There are four hardware threads, each of which renders a 64x64 tile at a time.  The program requires around 1.6 million instructions to complete, performs around 500k memory accesses, and uses 2.25 Mb of working RAM.