I had a look. There's nothing I can see that suggests they are centrally computing anything like 1920x1080x24 bits of data at 60Hz, that's 360MB of brand-new data per second per user, which is what our graphics cards are doing for us client-side.
I won't. And neither are they. Perhaps a very, very sparse approximation taking advantage of diagonal and/or block structure that actually computes a miniscule fraction of that.
