The big news here is actually the backstory. When you can use the GPU as generic, massively-parallel compute hardware (GPGPU) then it no longer matters if the host code is fast C/C++ or a high-level language like JS because all of the actual performance-critical work is happening in OpenGL/OpenCL/Cuda and the host is just orchestrating.
Video is on the cutting edge because it’s obviously parallel, but I predict other technologies will follow. Has someone done a pathfinding algorithm like A* in OpenCL? (If not today, then when hUMA becomes mainstream) That could allow a game with an absurd number of AI mobs intelligently navigating a dynamic battlefield. And you might be able to code it in the browser.
Originally shared by Ars Technica