1
edit
Changes
→Description
The new version of my fractal program is running on the GPU and has been highly parallelized. It is running more than ten times faster than the original version. I am sure this is highly GPU dependent and I am running it on an Nvidia Geforce GTA 660 Ti with 1344 cuda cores. However I think the comparison is fair because my CPU is also very fast. It is an Intel i5 2500k which is 4 cores at 3.3 Ghz stock and is running overclocked to 4.6 Ghz. A slower GPU compared to a slower CPU might be able to expect similar speed ups.
Parallelizing the main Mandelbrot function ended up requiring me to move a few other functions to the GPU and rewriting the anti-aliasing function. The anti-aliasing previously made many calls to the Mandelbrot function. This was not good for the GPU and I changed it so that it makes a list of points to run the anti-aliasing algorithm over. Another issue with the anti-aliasing was selection selecting the right value to return. I previously found the best quality result is to return the median value. The anti-aliasing required its own Mandelbrot function as well as moving the median function to the device. I also move the function that converts Mandelbrot escape values to RGB values to the device. This was because I found that I could account for half the running time depending on settings.
=== Code ===