Massively parallel GPU computing using the Nvidia tools is no harder than it needs to be. When I wrote GPU computing is about massive data parallelism way back in January, I hoped to have a CUDA-capable machine set up quickly. It took a bit longer than I thought, but I’m finally ready to tell you about the CUDA SDK.On Windows, the CUDA SDK integrates with Visual Studio. The documentation is good enough to get you started, although it wasn’t completely up to date when I did the download, and the installation has some manual steps. As you can see, when it’s all complete you get syntax-coloring for CUDA keywords and types. You can click this image to see a higher-resolution version.I’m looking at the Mandelbrot sample provided in the CUDA SDK; it was contributed by Mark Granger of NewTek. In fact, the SDK comes with a sample browser, in case you just want to play with the applications provided. Here’s the full Mandelbrot set. Note the frame rate:And here’s a zoomed-in section of the set, with double-precision turned on and the iterations bumped to 4096 for high detail:If we were running this with the CPU, that frame rate would not be in frames per second, it would be much slower, perhaps frames per hour. Color me impressed. Software Development