GPU accelerated computation (like CUDA/OpenCL) within the browser, using WebGL

learningwebgl.com

3 points·babs474··3 comments

What's the rationale behind doing this ? GPGPU stuff is used for the most part in pipelines of float intensive calculations, javascript is mostly used to implement user interface functionality in interactive websites, I can see that it is 'interesting' in an academic sort of way, just like the javascript based key parallel key breaker was interesting but I can't see anybody that is looking at production taking this as their preferred route to a solution.

Am I missing something ? Is there an application that would benefit from this (seti at home is mentioned, but that is wired in tightly with using idle cycles, information not generally available to the browser anyway, and it would be at considerable overhead compared to the C/C++ approach).

I can only guess about applications since this is pretty new, and only a proof of concept. I doubt anyone has had a chance to consider this for any production solution.

However, I think the possibilities could be pretty compelling for any distributed computing project that can be gpu accelerated. What if all you had to do to contribute to the seti at home project was to visit a webpage. No client to install. If the algorithm is sufficiently implementable on the GPU there would be little or no overhead compared to a C/C++ solution.

Historically javascript has been focused on user interfaces, but perhaps it has been overlooked as a distributed computing platform.

Easier scalable webapps, because it is possible to offload some data crunching from server (paid by us) to client computer (paid by him).

Examples?

Calculating and matching SURF points on large set of images for image matching, point-of-shot localisation and some more fun and profitable purposes.

Reply on news.ycombinator.com