| Dear HN, I've been working on some 2d visualizations in my free time, just for fun, things like cellular automata, chaotic iterated maps, fractals, etc. I've been using OCaml and its Graphics package, and sometimes Python with Cairo to do this -- I really just need a canvas where I can draw pixels. So far, I've been running all computations on the CPU, but I'd like to start playing with the GPU, since my frame rates are getting slow. I'm looking for some kind of GPGPU framework that's somehow integrated with a GUI library so that I can draw to the screen without shuffling memory CPU<->GPU too much (is this as big of an issue as I have convinced myself it is?). I know I can do this just writing shaders, but I'd rather not do that if possible. So far, I've looked at: - [SPOC][spoc] - seems ideal, but unstable - Tensorflow - I use it a lot, but can't find a way to draw to the screen without evaling to numpy array and drawing that - [Futhark][fh] - seems tough to draw to screen here too... Any tips would be appreciated! [spoc]: http://mathiasbourgoin.github.io/SPOC/ [fh]: https://futhark-lang.org/ |
the output of cuda will be in a texture map memory and given to opengl for rendering.
although I have not done anything with visualization using this approach, I have, however, did video processing with this approach. A video frame is first processed by cuda and then passed to opengl for rendering.